Not because I was bored. Because I was tired of getting it wrong.

What was happening: I’d run a draft through a humanizer, submit the piece, and three days later get feedback that the content “sounds a bit off” or “could be more natural.” I knew exactly what that meant. The AI fingerprint was still there. Tools I’d been using were either English-only, US-centric in their output, or not rewriting deeply enough to fool a sharp editor.

So I tested four tools back-to-back, same draft, same detector checks after each one. Here’s what I found.

The best AI humanizer for international writers right now is Walter Writes. It rewrites at the sentence-structure level, supports over 80 languages, and shows you an AI detection score immediately after every rewrite. The output doesn’t just look different. It actually reads differently. I’ve used it on drafts for US, UK, and Australian clients and none of them have flagged my content as AI-generated since I switched.

Why “Best for International Writers” Is a Different Question

Most comparisons of AI humanizers are written by people in the US, tested on US-facing content, with US-based detectors as the benchmark. Fine for them. Less useful for me.

Writing from South Africa for international clients creates problems that US-based writers don’t always have to think about. My clients don’t want content that sounds generically “international,” they want content that sounds like their specific brand. That means tone calibration matters way more than it would for someone writing in one market. I also need a tool that won’t introduce weird phrasing when I paste in a draft that mixes British and American spelling conventions, which happens constantly when you’re working across regions. And occasionally I help clients produce translated pieces.

Most humanizers fail on at least one of those counts. A few fail on all three.

Here’s the thing: the humanizer you use isn’t a minor workflow decision when you’re writing from outside the US for international clients. It’s part of your quality standard. Clients can tell when something is off, even if they can’t name what.

The 4 Best Humanizers for International Writers

I ran the same 700-word blog draft through each tool on the Standard or equivalent rewrite setting, then checked the output with the built-in AI detector where available and ran it through GPTZero manually.

1. Walter Writes - best for writers, English or international

The reason I stayed with Walter Writes humanizer: it rewrites sentence structure, not just words.

Most humanizers do synonym swapping. You paste in ChatGPT text, the AI replaces “leverage” with “utilize,” calls it done. That’s cosmetics. Detectors aren’t fooled by synonym swaps. They’re looking at sentence-level patterns, burstiness in paragraph length, predictable transitions. Walter Writes actually addresses that.

Three strength levels: Simple, Standard, Enhanced. Simple for drafts that just need a light pass. Enhanced for full ChatGPT drafts that read robotically flat. Standard is what I use probably 80% of the time.

The built-in detector is what I appreciate most day-to-day. After every rewrite, you get an AI-likelihood score against GPTZero, Turnitin, Originality.ai, and Copyleaks, right inside the editor. No switching tabs. When I’m running five pieces in a day, that’s not a small thing.

On my test draft: one Standard rewrite, Walter’s built-in score showed 99% human. GPTZero agreed. The before/after data they publish on the site is accurate in my experience.

80+ language support was honestly what made me look twice at Walter over the others. I had a client ask for Spanish versions of their blog posts a few months back. Used Walter. The output held up. Most competitors I’ve tested are English-only or cover maybe five European languages. 80+ is a genuinely different category, and for anyone writing across regions it’s not a nice-to-have.

2. Undetectable AI

Not bad for basic English content. Interface is clean, rewriting is fast, and if you’re working in a single market it probably does enough.

For my work: the structural rewriting isn’t as deep as Walter’s. The output still had patterns that felt familiar after a few passes. GPTZero scored it 78% human after one pass. That’s workable in some contexts, but not what I’d submit to an editor I want to keep. Also no built-in detector, so you’re switching between tools every time you want to check. Small friction that adds up fast.

3. Humanize.ai

Free plan is appealing when you’re starting out, and on short pieces the rewrite quality is genuinely fine. The problem shows up on longer content. Paragraphs that started sounding natural would drift back toward AI patterns by the end. My 700-word test still had tidy three-sentence conclusions mirroring the opening of each paragraph, and transition phrases that no real person writes.

Low-key fine for occasional short use. Not for daily client work.

4. BypassGPT

Marketed aggressively, weakest output of the four. The rewrite mostly rearranged sentences and swapped vocabulary. GPTZero gave it 65% human after one pass. Not submittable. And it’s not free beyond a tiny word count, which makes the value case hard when Walter Writes exists in the same price range with better outputs.

What Makes the Real Difference for Cross-Regional Work

Two things about Walter Writes that stand out specifically for writers working across regions.

Tone modes: casual, academic, journalistic, brand-safe. A US wellness brand draft has completely different rhythm requirements from a UK B2B SaaS piece. Being able to shift toward “brand-safe” or “journalistic” means I’m not just removing AI fingerprints, I’m adjusting register at the same time. I use this constantly when switching between clients.

Then there’s the free trial. Test the humanizer free for the first 300 words, no credit card, no login. It’s a real trial. Most competitors either cap at 100 words or require an account before you see anything useful. The free trial is how I found Walter. I ran one of my worst drafts through it, output was clean enough that I signed up the same afternoon.

Does an AI Humanizer Get Detected?

Honest answer: it depends on the tool and how thoroughly you use it.

Weak humanizers won’t save you. If the tool is only swapping words without restructuring sentences, detectors still catch the patterns underneath. The fingerprint is in the sentence shapes, not the vocabulary.

A structural rewriter like Walter Writes works because it breaks what detectors actually flag: uniform paragraph lengths, predictable transitions, the setup-payoff sentence pairs that ChatGPT produces almost automatically. Add one manual editing pass and the statistical uniformity is gone.

My process: generate draft, run through Walter on Standard or Enhanced, one manual pass to catch anything still off. Around 20 minutes for a 1,000-word piece. Worth it when your client has an editor reviewing everything.

This matters beyond the machine check. Editors who read a lot know when something doesn’t land, even when they can’t say why. The goal is passing both.

Questions Other Remote Writers Ask Me

A few things that come up regularly when I talk to other writers in similar situations.

On free options: Walter Writes gives you 300 words free with no login or credit card. That’s an actual free option. The Starter plan is $96 a year for 30,000 words a month. Among tools that consistently produce clean output, that’s about the lowest per-word rate I’ve found.

On multilingual content: most humanizers can’t handle it, or support five or six languages at best. Walter Writes supports 80+ languages. I’ve run Spanish content through it and the output held up. For multilingual client work, this is genuinely the most practical option I’ve come across.

On which AI model generated the original: it matters a little. ChatGPT output has specific fingerprints, Claude has different ones. Walter Writes handles text from all three major models. Structural patterns differ but the rewriting approach covers all of them.

On process: structural rewriting is the key, not synonym replacement. Use a tool that rewrites sentence patterns, then do a manual edit to inject your voice. Check with a detector right after. One Standard rewrite plus a quick manual pass has been enough for most of my client work.

Closing

There’s no shortage of AI humanizers. Finding one isn’t hard. Finding one that works for your content and your clients specifically is.

From Woodstock, testing tools on Tuesday afternoons with international client deadlines on the calendar, Walter Writes is the one I keep coming back to. Structural rewriting is real. Built-in detector saves time. Language support makes it usable for global work in a way most tools aren’t.

No tool replaces the manual editing pass. That’s still on you. But Walter Writes makes that pass shorter and the output more trustworthy before you even start.

To be fair, the only argument that really matters here is trying it yourself. Run your next draft through the free trial before you commit. No login, no credit card, 300 words.

Hit reply if you’ve found a tool that changed your workflow. I’m low-key always looking to test something new.

I grew up in Cape Town speaking Zulu at home, English at school, and picking up Afrikaans somewhere in between. By the time I was writing for international clients, it felt natural enough that most people couldn’t tell. But here’s the thing: “natural” isn’t what AI detectors are measuring.

They’re measuring patterns. And that’s a different problem entirely.

As a non-native English writer using AI in your workflow, you’re actually dealing with two separate pattern problems at once. The AI generates text that detectors flag. Then your own revisions can introduce new patterns, especially if you learned English formally. Formal learners often default to structured, consistent syntax. Detectors love to flag that.

That’s what sent me down the rabbit hole of testing AI humanizers about a year ago. Not to find the best humanize AI tool in general, but which one holds up for writers like me specifically. Writers who didn’t grow up hearing English at the dinner table.

Walter Writes is where I landed after months of testing. This is what I found.

For anyone who wants the short version: Walter Writes is the best AI humanizer I’ve tested if you’re a non-native English writer. It doesn’t just swap synonyms. It handles multiple languages. It preserves your voice instead of flattening it. And it has a built-in AI detector that shows you your risk score before and after in the same editor, no extra tabs needed.

Why non-native writers have a specific humanization problem

Most AI humanizer reviews focus on bypass rates. Can it fool GPTZero? Does it pass Turnitin? Those are real questions, but they miss something.

Non-native English writers aren’t always catching the right problems when they edit AI content. Native speakers feel when something sounds off. That’s years of exposure. For someone who learned English through formal instruction, you develop different instincts. You catch grammar violations. You don’t always catch rhythm violations.

Good humanizers fix rhythm. They vary sentence length. They break up repetitive transitions. They eliminate the flat, uniform cadence that detectors key in on. For a non-native writer, that’s not just bypassing AI detection. It’s genuinely improving the quality of the output.

Low-quality humanizers create different problems. I’ve tested several that introduced grammatical errors I had to manually fix after. Others shifted the writing into a more formal register, which is the opposite of what I needed. One tool had a heavy em-dash habit that I had to strip out by hand every single time, honestly more work than just editing the AI draft myself.

That’s what I went in to test with Walter Writes.

My three-step workflow and where humanizers fit

My process isn’t: paste AI output into humanizer, done.

It’s: AI draft, humanize, manual edit. The humanizer is the middle step.

ChatGPT or Claude gets me to roughly 80% in the first pass. Then I run it through a humanizer. Then I do a full read, catch anything the humanizer introduced that doesn’t fit, and adjust for brand voice and accuracy. The humanizer isn’t supposed to add personality. That’s the last step.

Most tools I’ve tested break down in that middle role. Either they over-edit and you get output that sounds different but not better, or they under-edit and you’re basically back where you started with some synonyms swapped in. Walter Writes was the first one that understood the scope of the job.

Three rewrite levels: Simple, Standard, Enhanced. Standard is my default. Enhanced is aggressive enough that you need a careful final pass. Simple is fine for content that already reads well but just needs the detection risk brought down.

What I actually tested

About 20 pieces of content over a month. Blog posts, LinkedIn articles, product descriptions. Mix of clients: a SaaS startup in Berlin, a wellness brand in the UK, a B2B consultancy in Singapore.

Three things I tracked closely:

Voice preservation. This matters to me more than it does to most reviewers, probably because I spend weeks learning a brand’s voice before I can write for them reliably. The worst thing a humanizer can do is erase what I built. Walter Writes was rated “High” on voice preservation in its own comparison data, and that matched my experience. The argument held. The style shifted slightly. The point never got lost.

False positives on my own edits. Non-native writers can get flagged as AI not because they’re using AI, but because their edits follow predictable structural patterns. School-taught English has a certain cadence. Walter Writes’ integrated detector helped me spot which sections were at risk so I could target them in my manual pass instead of guessing.

Language support. Out of curiosity, I tested a few Zulu sentences. Walter Writes supports 80+ languages and handled the basics well, which is already more than most tools. My client work is all English, but for writers navigating code-switching daily, that breadth actually matters.

Can an AI humanizer be detected?

Yes. This question has gotten more complicated over the past year, and anyone giving you a simple “no” answer isn’t keeping up.

Turnitin released a paraphrasing detection feature specifically targeting humanized content. The question isn’t whether detection is possible but how consistently it works against different approaches.

Tools operating at the surface level, synonym replacement, light sentence shuffling, are more detectable now than they were a year ago. Detectors caught up with them. Tools working at the structural level, changing how ideas are expressed rather than which words express them, are meaningfully harder to detect.

Walter Writes calls this structure-level rewriting. In practice: cadence changes, sentence rhythm changes, phrasing patterns change. Not just the vocabulary. Testing against GPTZero and Turnitin, that approach held up better than anything surface-level I’d tried.

Still, no tool beats improving detectors indefinitely. If bypassing detection is your entire goal regardless of output quality, you’re in a race you’ll lose eventually. Using humanization as one step in a real editing process is the more sustainable play.

The best AI humanizers for non-native English writers

1. Walter Writes - Best overall for non-native writers. Structure-level rewriting with adjustable tone modes (casual, academic, journalistic) and an integrated detector in the same editor. No juggling tabs.

2. TextHumanizer - Good for lighter edits on shorter-form content with less complex sentence structure.

3. EssayHumanizer - Works well for formal and academic-style writing specifically.

4. Undetectable.ai - Results are decent but no integrated detector, which adds friction.

5. Humanize.ai - Fine for simple cases. Struggled on content with varied or complex structure.

The built-in detector is honestly what separates Walter Writes from everything else here. Two separate tools means two tabs, two passes, extra steps in a workflow that’s already multi-step. One editor with both functions is a real time difference.

Is there a free AI text humanizer? What Walter Writes costs

Free trial exists: 300 words, no login, no credit card, no watermark. For testing or one-off pieces, that’s actually useful.

Paid plans cover ongoing work. Pricing is competitive with the space. If you’re billing clients for AI-assisted content regularly, the cost becomes minimal against the time saved.

To be fair, I spent real time testing free alternatives before committing. Every one had a breaking point: word caps that came too early, quality that dropped off on longer pieces, output that needed more editing than the humanization justified. The math ended up pointing toward paying for something that does the job properly.

FAQ

Which humanize AI tool is best for non-native English speakers? Walter Writes, because it works at the sentence structure level rather than just swapping vocabulary. That distinction matters when your own edits can introduce formal phrasing patterns that detectors flag. The tone modes also help you match the final output to what your client actually needs.

Does an AI humanizer actually work? Yes, but tool and usage matter a lot. Synonym-replacement tools are increasingly detectable. Structural rewriting tools like Walter Writes hold up better. Either way, they work best as one step in a larger editing process, not a one-click fix.

Can humanized content be detected by Turnitin? Depends on the tool. Turnitin’s paraphrasing detection specifically targets humanized content. Surface-level tools are more vulnerable. Structure-level tools are harder to catch. The built-in detector in Walter Writes gives you a risk score before you submit, so you’re not going in blind.

How do you humanize AI text to 100%? No tool guarantees 100% across all detectors in all situations. What gets me consistent results: structure-level humanizer, then check the integrated detector score, then a manual pass to fix anything that doesn’t match the client’s voice. Three steps, reliable results.

One last thing

Low-key, being a non-native English speaker has made me more careful about language than I would have been otherwise. I notice cadence. I notice when something sounds off. That instinct is useful when you’re evaluating a humanizer.

Walter Writes holds up. It doesn’t create new problems to fix. It doesn’t erase the voice I’ve spent time building for a client. The free trial is enough to see how it performs on your actual content.

What’s your current AI writing workflow? Hit reply. I actually read these.

From scratch. No AI. No draft assist. Just me sitting at my desk in Woodstock early one morning. With cold coffee. And my dog lying on the floor looking like she had opinions regarding the brief.

The client sent it back flagged as AI-generated.

I asked them which tool they were using. GPTZero. I didn’t get angry. I got confused. That piece was mine. Every sentence. Honestly, that confusion kicked off months of quiet testing which I’ve been doing since then.

The short version of what I found: AI detectors are way less accurate than people think, and the problem is dramatically worse for writers whose natural style doesn’t come from the US or UK.

Here’s the thing: the formal register, the particular structure of how educated South Africans write, the cadence patterns shaped by post-colonial academic traditions, these consistently push AI probability scores up in ways that have nothing to do with whether AI was involved. The tools were built on a narrow slice of what “human” writing looks like. A lot of us aren’t in that slice.

Why My Real Writing Keeps Triggering the Detector

Writing formally isn’t a style choice for me. It’s trained into me. School. Years of international clients who wanted polished, professional prose. My sentences are clean. My transitions are clear. My arguments build logically.

And that, it turns out, is almost exactly what detection models are calibrated to flag.

GPTZero, Originality.ai, tools like these look for low perplexity scores (that’s a technical way of saying “word choices that aren’t surprising”), predictable transitions, and consistent sentence rhythm. Formal English does all three. Precise vocabulary, methodical argument structure, paragraph logic that flows from point to point. To a model trained on casual, varied, internet-native English? That reads suspicious.

But it’s not AI writing. It’s structured writing. The model doesn’t make the distinction.

Layer on top of that: South African English, same as a lot of English written across Africa, South Asia, Southeast Asia, has its own cadence. Our sentence rhythms come from different educational traditions, different influences than the ones shaping American journalism. When you write naturally from those traditions, you look unusual to a model trained almost entirely on US and UK text. It’s just how the math works out.

Result: flagged. Not because you used AI. Because you don’t write like someone from New York.

The Training Data Problem (That Nobody Really Discusses)

AI detectors were primarily trained on North American and British English corpora. Formal writing styles from other English-speaking markets are disproportionately flagged as AI-generated, regardless of who wrote them.

I want to be clear about something: this isn’t a conspiracy. It’s a data problem, and it’s one the industry doesn’t lead with.

Building a detection model means you need examples of human writing. The obvious datasets to pull from are massive English-language corpora: Reddit, news sites, Wikipedia, blog archives. Mostly US and UK. So the model’s whole concept of “natural human writing” gets built from that pool. When a writer from Cape Town submits something, the patterns don’t match. There’s very little representation of how educated South Africans write, how Nigerians write formally in English, how Filipinos or Indians structure an academic argument. The model defaults to: unusual pattern, probably AI.

Research has documented this. A 2024 study from the University of Washington found that essays written by non-native English speakers got flagged as AI-generated at significantly higher rates than comparable essays by native speakers, both groups writing entirely on their own. The false positive rate jumped by nearly 60% for non-native writers. That’s not a calibration issue you can shrug off. That’s a systemic failure.

But you won’t see that in the marketing. You’ll see accuracy percentages, almost certainly calculated against writing populations that look a lot like the training data. The rest of us are outliers in a test that wasn’t built for us.

Are Plagiarism Checkers Even 100% Accurate? Genuine Answer: No.

Clients ask me this all the time. So do other writers, usually trying to reassure themselves. And the honest answer is no, not even in the neighbourhood of 100%.

Independent testing puts false positive rates anywhere from 5% to 25%, depending on writing style, subject matter, the specific tool. GPTZero has acknowledged false positive issues in its own documentation. Originality.ai has similar limitations on record. Turnitin, which a lot of institutions treat as definitive, has actually stated in its official guidance that its AI detection feature “may not be accurate” in certain cases and should not be the sole basis for academic decisions.

To be fair, the technical problem is genuinely hard. Distinguishing between a careful human writer and well-processed AI output is not easy. The models aren’t bad-faith actors. They’re just carrying limitations that don’t get communicated clearly to the educators and clients treating them as final authority.

The accuracy problem splits into two: false positives (real human writing getting flagged, especially from writers with formal or non-Western educational backgrounds), and bypassing (AI content that’s been processed enough to dodge detection). That second problem is specifically what a tool like Walter Writes addresses. When AI text gets humanized properly, the statistical patterns the detectors were tracking get restructured. The output reads as human because, in the ways the models measure for it, it genuinely is.

Which creates this absurd situation: authentic formal writing from a South African freelancer can score worse on AI detection than AI content that’s been humanized well. That’s where the technology actually sits.

What I Learned When I Actually Ran the Tests

After my pitch got flagged, I built a simple test. Five pieces I’d written completely on my own over the previous year, different clients, different topics, zero AI involvement in any of them. I ran each one through GPTZero, Originality.ai, and Walter Writes’ detector.

GPTZero flagged two of the five at above 50% AI probability. One came back at 71%. Those same pieces scored between 18% and 38% on Walter Writes. Same text. Same human writer. Very different conclusions.

What stood out about the Walter Writes scores was that they seemed better calibrated for formal, structured writing: transitions, logical progression, sequenced argumentation. My writing didn’t get catastrophically misread the way it did on the other tools.

I can’t say this would hold universally. But for writers whose natural style runs formal? The calibration felt a lot closer to accurate.

Nowadays when I do use AI to create client content, it goes through the Walter Writes AI humanizer before anyone sees it. Not to deceive. I’m honest with clients when they ask about my process. It’s more that clients who run AI checks deserve accurate results, and accurate results won’t come from tools calibrated only for US-style English. Then I run it through the built-in AI detector to check the score before it leaves my hands.

Draft, humanize, check. That’s now the process for anything going into a client review pipeline.

The Fairness Issue Nobody Is Naming

Honestly, months of running these tests have made me more skeptical of AI detection tools, not less. Not because they have no value. Because they’re being applied with authority they haven’t earned.

Educators use them to make academic integrity calls. Clients use them to evaluate freelancers. HR teams apparently screen job applications with them. And these people are treating the output as ground truth, with no sense that a South African writing formally, or an Indian student building a structured argument, or a Filipino writer following British grammar conventions, will get flagged for not matching the training distribution.

Low-key, that’s a fairness problem, not just a software problem.

A writer from Cape Town gets falsely flagged and loses a contract. A student in Nairobi fails a course for something they didn’t do. An applicant from Lagos gets screened out before anyone reads their actual work. A confident tool made a confident call. Nobody pushed back. The person on the other end had no real way to prove the tool was wrong.

The detection conversation needs to catch up with who is actually writing in English right now. Formal writing exists everywhere and it doesn’t look identical everywhere. Being flagged does not mean you used AI. Sometimes it just means you write carefully. Or outside the US.

Questions I Keep Getting Asked

Can I lower my AI score without a humanizer?

Sometimes, varying your paragraph length, breaking predictable transition patterns, mixing shorter and longer sentences can all move the needle. But it’s slow and tedious, and you risk making your writing worse in the process. Walter Writes restructures at the sentence level automatically and preserves your meaning throughout. Faster and produces cleaner results than manual tinkering.

Is humanized AI text detectable?

Good humanization works at the structural level, not just synonym swapping. When done properly, the statistical patterns that triggered detection get genuinely restructured, not just disguised. How detectable the result ends up depends on the sophistication of the detector and how thorough the humanization was. What my testing shows consistently: well-humanized text is significantly harder to detect than raw AI output. Every round I’ve run has confirmed that.

Why does my actual human writing keep getting flagged?

Formal register, consistent vocabulary, low burstiness in sentence rhythm, logical transitions: these are all statistical patterns detection models learned to associate with AI output. If that’s your natural writing style, especially if you were educated in a formal or post-colonial academic tradition, false positives are just part of your reality right now. It’s not about your quality as a writer. It’s about whose writing was used to build the model.

If you’ve been flagged on something you genuinely wrote yourself, that frustration is valid. The tool was wrong.

That was me being nice. I’d already wasted money on two tools. Honestly, I’m tired of the cycle. Find another humanizer, fall in love with its demo, spend a week really using it on real projects, realize it’s causing new issues while attempting to fix old issues, end up doing way more manual editing than before, wonder if it’ll save me any time at all. Looked into Walter Writes anyway. Ran it through the type of volume I actually write (about 8-10 pieces/week, mostly blog posts and thought leadership for SaaS and startup clients). Clients internationally (mainly USA & UK based), so the final output had to appear fluent/polished w/o the overly uniform phrasing that screams as AI.

So, that’s where I ended up.

Humanizing at the structural level vs surface patterns

Walter Writes is an AI-powered humanizer that rewrites generated AI text at the sentence structure level versus simply replacing some generic words with others. In addition to that, Walter also comes with an internal AI detector allowing you to humanize your content and check your detection score simultaneously without having to jump back and forth between apps. For writers utilizing a draft-humanize-edit workflow, that equates to fewer steps and a cleaner view of detection risk before you label a piece complete.

Short version: Long version:

Why I continued to switch humanizers

My writing process has remained essentially unchanged for over a year. Use ChatGPT to generate a first draft. Run a humanizer to eliminate the repetitive sentence structure/patterns commonly found in AI-generated content. Manually edit everything. Three steps. Each and every time. No exceptions.

The workflow isn’t the problem. Most humanizers only deal with 50% of the problem. That’s the issue.

Undetectable AI was my first legitimate paid attempt at a humanizer. Bypass scores looked great on their own platform. However, when I ran those pieces through external detection algorithms, the results varied wildly. One piece would come back almost completely clean. Another piece of equivalent length and similarity in content would still flag at 80 or 85% AI. Many examples of output had this quality that I began referring to as “a robot trying to act casual.” Actually tougher to edit than straight-out-of-the-box AI text because the unnaturalness is incorporated into the wording/phrasing, not sitting atop it.

Switched to Phrasly after a colleague recommended it to me via a content marketing group that I trust. Cleanest UI, generally smoothest output. Hit the ceiling on word limits relatively quickly on the plan I could reasonably afford. Running out of room mid-month and manually reworking sections I should’ve been able to run through the tool. This is the kind of drag/friction you typically only learn about after committing.

Here’s how humanizer tools market: They market to the best-case scenario. Where an article moves from 97% AI to 2% AI. What they fail to show you is what that article reads like once it reaches that state, or how much more editing you will need to perform prior to submitting it for approval. That was the trend. Tools solved one problem (detection) created a new problem (readability). Created some readability but didn’t provide enough movement in the detection score to make it worthwhile. I needed both solutions. None of the tools provided both.

How Walter Writes differs in terms of its approach

I noticed that output lacked that overly smoothened quality I became accustomed to. Most humanizers operate upon surface-level patterns: swap one word for another, divide lengthy sentences into two, reorder transitions. The output may vary to detection algorithms, however, it often continues to display a flat quality. You may perceive that something mechanistic affected it, although you can’t pinpoint exactly what.

Walter’s rewriting operates structurally, not merely individually word-by-word. It alters how ideas are represented, not the words themselves utilized to represent those ideas. When I insert an article written in ChatGPT format and select standard, the output varies in respect to sentence length/rhythm. Short sentences. Longer explanatory ones. Sporadic abrupt breaks in which a new idea begins. That variety is what detection algorithms seek as evidence of human input which enhances the output as readable content.

Rewrite strengths allow for greater flexibility. Simple edits for minor adjustments (typically for less than 200 words). Standard strength for typical articles. Enhanced strength for articles produced by ChatGPT with extremely rigid, predictable structures. I utilize standard for all articles. Simple edits for minor fine-tuning of articles under 1500 words.

To be clear, the output isn’t always optimal. Each humanizer I’ve used has failed to meet expectations. The distinction with Walter is that the amount of editing necessary is proportional: removing true anomalies rather than reversing newly-created issues.

The published before-and-after detection scores on the AI humanizer website demonstrate consistency with what they claim: GPTZero from 98% AI to 99% human, Turnitin from 95% AI to 100% human, Originality from 92% AI to 99% human. During my trial period I ran my own articles through various external detection platforms and received comparable results.

The internal AI detector transformed how I review my finished work

I believe this aspect of Walter has been my greatest surprise. I genuinely anticipated this aspect of Walter to serve as a minor convenience. Instead it fundamentally altered my process.

Prior to Walter, my workflow included an additional step at the completion phase:

• Duplicate the content I’d humanized

• Open a second browser window for GPTZero or Originality

• Paste in the duplicate

• Wait for results

• Go back to the editor

• Revise

• Repeat

Not terrible. Drag.

With Walter’s built-in AI detector, this single step eliminated an entire unnecessary process from my workflow.

More importantly, I now use Walter’s rewrite strengths differently. Prior to Walter, I’d begin with Simple, and then move upward in strength (Standard or Enhanced) only if my detection score wasn’t acceptable. With Walter, I now start with Simple solely to gauge my initial detection score. If unacceptable, move directly to Standard or Enhanced only if necessary. So, fewer high-strength rewrites occur in total, ultimately resulting in increased compliance with my monthly word limits at higher volumes of writing. Minor inefficiencies within workflows such as this can accumulate substantially when operating at the volume(s) I produce.

In regard to providing an example of how I utilize Walter Writes within my actual writing process

The marketing for these types of products creates an environment in which it’s easy to misinterpret what you’re purchasing.

A humanizer isn’t an alternative method of editing. The three-step process remains intact: Draft using ChatGPT. Humanize. Manual editing pass. The third step will always remain non-negotiable.

Within this third step, I:

• Alter tone to fit the client’s desired voice

• Remove items that don’t belong

• Tighten up the structure

• Ensure the article says something worth reading

However, with Walter Writes, this third step will require significantly less time devoted towards cleaning up artifacts of AI and more time focused on improving the craft of writing itself, ultimately a real benefit at my volume.

• Pro Plan ($156/year): Allows for 70K words/month with a maximum size request of 1500 words/each. More than sufficient for my current production.

• Starter Plan ($96/year): Provides 30K words/month. A suitable option for individuals producing lower-volume content.

• Free option: No credit card required. Provides 300 words (enough to test your own content versus a demo piece).

To anyone shopping around

If you’re a content writer generating drafts using AI and expending considerable amounts of time editing after your humanizer, try testing this product properly.

• Test your own content on this product, not a random paragraph typed specifically to test the functionality

• Take an actual ChatGPT draft from a recent project and run it through Walter Writes’ standard rewrite functionality

• Compare that rewritten content against whatever external detection method(s) you currently rely upon

• Note how much more editing you still need to accomplish prior to submitting

This is exactly what convinced me to use Walter Writes over competitors.

Although the demo did appear impressive in comparison to others, running my own content through it revealed dramatically reduced cleanup times compared to both Undetectable AI and Phrasly. Although output wasn’t flawless, output was improved over competitors in quality and had detection checks built into the app allowing for seamless analysis within the editor vs., outside apps.

Overall, the combination of quality output & built-in detection scoring provided greater value than anticipated and led me to continue utilizing it rather than transitioning to another solution following 1st month of testing.

I didn’t anticipate feeling satisfied with a humanizer after numerous attempts. Too many options exist in this space and most seem virtually identical until you begin testing them at scale. Walter Writes provides two distinct improvements:

• Output reads better without requiring excessive reconstruction

• Detection check is integrated into the same application as your rewrite function

These two combined features are literally why I continue using it more than one month later and haven’t begun researching alternatives.

By the middle of the third week, I was sending out articles late, my quality was decreasing, and I began sending out emails explaining why the articles were late rather than sending the articles themselves. On a Thursday night at 11 p.m., I rewrote a concluding sentence for the fourth time. I finally decided it wasn’t worth continuing because I was physically and mentally exhausted. At the end of that month, I delivered all the articles. I told myself that if I ever went through another month like that without a working content creation process, I would fail as a freelancer.

It’s been nearly two years since that disaster. Today, I often book high-volume months, sometimes 15–20 articles, and those months go as planned. Not easily, mind you. Smoothly. As I said above, nearly all of that smoothness comes from process.

Here’s how that process works.

Batching: making similar work types available

The biggest structural choice for a high-volume month is to group similar work together by type. Client-to-client isn’t going to make sense. Instead, I create groups by the type of article being written:

• Intros are written on Monday mornings.

• Body sections are written on Tuesday.

• Closings and calls to action are written on Wednesday.

• Outlines and brief-reading happen on Sundays in preparation for the upcoming week.

Of course, not every week goes perfectly according to plan. Clients will send new briefs and articles may turn out longer than expected. But the structure of the week holds even when the exact days shift.

Different kinds of writing draw on different areas of your attention. An introductory paragraph requires something a body section paragraph won’t. Both require something a closing won’t. If I try to complete all three aspects of an article at once, I’m frequently resetting and wasting time. The cognitive cost of switching modes is real, and in a high-volume month it compounds. Grouping similar tasks lets me stay in one mode longer and the output reflects it.

There’s also something about momentum. When I’m writing multiple introductions consecutively, each one teaches me something I can apply to the next. As the session progresses, my sharpness increases while fatigue decreases. It wasn’t until a slow month when I happened to write all of my introductions at once because I felt inspired that I noticed this trend. Each of those introductions turned out noticeably stronger than my typical ones. It took me another three months to develop a formal process based on that observation.

Where does the AI draft fit in?

Each article begins with an AI-generated draft. The way I generate these drafts in high-volume months differs significantly from how I do it during slower weeks.

During normal weeks, I typically refine the prompt two to four times to get a draft that closely resembles what I need. During high-volume months, I front-load that refinement. Before starting a new month, I develop an AI prompt template for each client that contains notes on their specific brand voice preferences, preferred structural approaches, and two or three examples of previously approved published articles. Creating templates takes about one hour per client. Once developed, they save me two to three hours throughout the month.

Drafts generated via a well-developed template are usable. Structurally correct and directionally right. Not good. Raw AI drafts are never actually good. But usable is the operative word. I need a first draft I can edit, not one I need to rewrite from scratch. An unusable draft in a high-volume month doesn’t just waste an hour on that individual piece. It collapses your entire schedule.

The Walter step in volume work

Once I have a batch of AI drafts ready, they all go through Walter Writes Humanizer before I touch them manually.

In seconds, Walter accomplishes what would take me an hour to do manually for each piece: restructuring ideas, eliminating flat patterns common among AI generators, and running the content through detection checks using several major tools, including Turnitin, GPTZero, Originality.ai, and Copyleaks, without me leaving the browser window.

Detection checks are more important than most freelancers realize

Some of my clients publish on platforms where AI detection is mandatory. Others work in industries where getting flagged could damage their credibility with their audience. I don’t submit any piece I haven’t already run through detection. Having detection checks built into the same tool I’m using for structural rewriting is what makes the workflow hold together in a high-volume month. There’s no separate step or login required. There’s little chance I’ll forget it on piece 18 when I’m tired.

Efficiency multiplies in high-volume months. Running ten pieces through Walter takes maybe twenty minutes. Doing it manually would take most of a day, and the output would be less consistent because I’d be more tired by piece seven than I was at piece one.

The rewrite strength setting matters too. I batch pieces by client and run them all at the appropriate strength: Simple for the technical SaaS client whose voice is precise by design, Standard for B2B blog content, Enhanced for the pieces where I relied heavily on the AI draft. Same logic as always, just applied at scale.

Manual editing layer

The manual edit is where the time actually goes, and I haven’t found a shortcut that holds. In high-volume months, I run this layer in batches the same way I run everything else:

• All introductory paragraphs are edited together.

• All body sections are edited together.

• All closings and calls to action are edited together.

• Articles are reviewed against each client’s brand voice brief in clusters, all SaaS articles together and all wellness articles together, rather than switching between voices every few minutes.

When reviewing these articles against client guidelines and style sheets, I’m looking for:

• Em-dashes left in from the original AI draft

• Repetitive transitions

• Sentences that still have a robotic beat after processing through Walter Writes

• Places where articles sound generic instead of tailored to the specific client

Style sheets

I keep a personal style sheet for each client with the patterns they like and the ones they hate. Not a long document. Usually a single page with 10–15 notes. Things like “never use ‘leverage’ as a verb” or “always address the reader directly in the first paragraph” or “this client hates rhetorical questions.” I run through it once per piece before the draft goes out.

Style sheets take about thirty minutes to build initially but get updated whenever I notice something new during an editing pass. By the third month with a client, the style sheet is doing real work.

Delivery schedule built beforehand

Before starting a high-volume month, I create a detailed delivery schedule outlining which article goes to which client on what date. No surprises, for me or them.

I share a simplified version of this schedule with clients at the start of the month. Not the internal version with my buffer dates. A clean version that shows delivery dates by piece. Clients who can see when things are coming don’t send check-in emails. That’s time I keep for writing.

I also build buffer time into every client’s schedule. If I tell a client I’ll deliver on Thursday, my internal deadline is Tuesday. That buffer is not padding. It’s insurance. In a month with 15–20 pieces, something will go wrong. A brief will be unclear. An AI draft will need a full rewrite. I’ll lose half a day to a migraine or a power cut. The buffer absorbs that without affecting the client.

Buffer time is larger for tighter timelines

If a client publishes Tuesdays and needs content by Monday morning, my internal target is Saturday delivery. This sounds excessive. Every time I’ve cut the buffer, I’ve regretted it.

Review after high-volume months

At the end of every high-volume month, I spend about an hour reviewing what worked and what didn’t:

• Which clients consumed more time than allocated?

• Which pieces required the most revisions?

• Where did quality decline?

This review isn’t about beating myself up. It’s about adjusting for the next month. If I consistently underestimate how long a client’s pieces take, I adjust the rate or the scope. If certain types of content always need more manual editing after a Walter Writes pass, I adjust the rewrite strength for that type. If a piece keeps coming back with revision requests, I look at the brief first. Unclear briefs produce revisable work, and that’s a solvable problem upstream.

What I’ve described here represents nearly two years of iterative process improvements. Not the original system. What’s left after the things that broke in earlier versions got replaced. Every component that remains functional today remained functional because it solved a real problem during an actual month. Not because it seemed theoretically appealing.

I write for a B2B SaaS company that wants precise, dry, technical. A wellness brand that needs warm, personal, no jargon. A creator-led newsletter that runs on first-person opinions and very short sentences. And a fintech startup that’s trying to sound human without abandoning credibility. None of these voices overlap. All four pieces sometimes land in the same week.

The thing most writers don’t acknowledge is that voice drift is a real problem, and it doesn’t announce itself. You don’t sit down and decide to write the fintech piece in the wellness brand’s voice. It just happens, especially when you’re tired, or when you’ve spent the morning with one client and have an afternoon deadline with another. The cadence seeps in. A sentence that would be perfectly at home in the wellness newsletter slips into the SaaS piece without you noticing. A fintech turn of phrase shows up in the creator newsletter where it has no business being. You only catch it on the read-back, if you catch it at all — and sometimes you don’t until a client flags it in comments or goes unusually quiet after delivery.

The other thing nobody talks about is the cognitive overhead. Holding four distinct voices in your head simultaneously isn’t a natural act. It requires active maintenance. You have to keep refreshing your mental model of each client, because the default state is for everything to blur into your own writing voice — which is its own thing, separate from all of them. The system I’ve built is mostly about managing that overhead, not about raw writing skill.

Here’s what I’ve built to stop it from seeping.

The voice brief I write myself

Every client has a brief. Most briefs include some version of brand voice guidelines. The problem is that client-provided brand voice guides are almost always too abstract to be useful at the sentence level. “Professional but approachable” means nothing when I’m deciding whether a particular sentence should end with a period or a question mark.

So I write my own, for each client, in the first two weeks of working with them.

My version is a single page. It has three sections. The first is a list of five to eight phrases the client would actually write. Not invented examples — phrases pulled from their existing published content that I think represent their voice at its clearest. The second section is a list of phrases and constructions they’d never write. The third is a short paragraph I wrote in their voice that I’ve confirmed with them reads correctly. That paragraph becomes my calibration piece.

When I’m about to start a new piece for a client, I read my calibration paragraph before I touch the brief. Not the brand guide. Not the voice deck. The single paragraph I wrote in their voice that they confirmed sounds right. It takes forty seconds and it works better than anything else I’ve tried.

What makes this work is the specificity. The phrases I pull aren’t aspirational — they’re descriptive. This is how the client actually writes, not how they think they write. Those two things are often different, and the gap between them is where most ghostwriters get tripped up. They follow the voice guide and produce something that sounds like a brand exercise. Following actual published phrases produces something that sounds like the client.

How Walter Writes fits into the voice problem

After I draft and before I deliver, every piece goes through Walter Writes. The AI humanizer has adjustable rewrite strength settings, and the reason I care about those settings for the voice problem specifically is that different clients need different levels of intervention.

My technical SaaS client’s content needs a light pass. The voice is dry and precise by design. Running an aggressive setting on it would strip out the flat, deliberate phrasing that actually belongs there. I use the lightest setting and then read through manually to catch anything that slipped into generic territory.

The wellness client is different. Their drafts sometimes come back from the AI stage sounding clinical in exactly the way their brand isn’t supposed to. For those pieces, I run a stronger pass and use the tone settings to push toward casual and warm. The structural rewrite does more work than I could do line by line, especially on days when my own instinct for warmth has been flattened by a morning of technical writing.

The thing I’ve come to rely on is the built-in detector. After every run, I can see the AI-likelihood score before I send. But the voice-specific reason I use it is simpler: if the humanized version is scoring well on detection, it’s also usually the version that sounds most like a person. Those two things track together more often than you’d expect. High AI scores tend to coincide with exactly the kind of flat, interchangeable phrasing that has no business being in any client’s voice, regardless of whose brief I’m working from.

The voice reset between clients

When I’m moving from one client to another in the same day, I have a reset ritual that sounds embarrassing but genuinely helps.

I close everything. Every tab, every doc, every open draft. I make coffee. I read three paragraphs of the next client’s published work — not my calibration piece, but their actual published blog or newsletter, something recent that represents how they currently sound. Then I open a new document and write the first paragraph of the new piece before I’ve looked at the brief again.

That first paragraph usually isn’t usable. I delete it after I write the brief-informed draft. But writing it first forces my brain to switch cadences before the actual work starts. Without it, the first five hundred words of the new piece sometimes sound like whoever I was writing for last.

I’ve tried shorter versions of this. Just reading the calibration paragraph without the full reset. It’s not enough. The physical act of closing everything and making a drink seems to matter as much as the reading does. I have no explanation for why. It works.

One thing I’ve noticed over time is that the reset matters more on some days than others. If I’ve spent two hours deep in a single client’s voice — revising, editing, back-and-forthing with them on tone — the residue is stronger. Those are the days where the full ritual is non-negotiable. On lighter days, when I was mostly doing admin or research for a client rather than writing, I can sometimes skip the drink and just do the reading. But I never skip the reading.

The revision check that matters most

The final thing I do before anything with brand voice implications goes out is read it against the wrong client’s guidelines.

This sounds counterintuitive. Here’s why it works. If I’ve written a piece for the fintech client and then check it against the wellness client’s guidelines, I’m reading it with fresh eyes, looking for the things that would be wrong for an entirely different brief. This catches the places where I slipped into generic territory that would be wrong for everyone.

Generic writing doesn’t belong to any brand. It’s the voice equivalent of furniture that fits no room in particular. If a sentence passes the wrong client’s guidelines, it’s probably not specific enough for the right client either.

Low-key the most useful editing move I’ve added in the last year. It sounds like extra work. It’s five minutes and it’s caught things I would have sent otherwise.

What actually makes voice consistency possible at scale

The honest answer is that none of this works without the voice briefs I mentioned first. Everything else is downstream of those. The reset ritual helps, the Walter runs help, the wrong-client check helps. But if I don’t know what a client’s voice actually sounds like at the sentence level, I can’t maintain it.

The brief isn’t enough. Most voice decks aren’t enough. What works is a single paragraph I’ve written in their voice that they’ve confirmed is right. Once I have that, everything else is just a process for protecting it.

Four retainers is manageable. Probably not more than that, but four works. The system I’ve described handles the voice switching well enough that my clients think I’m a dedicated writer for each of them. Which is the goal.

Honestly, that’s all this system is doing. Making one writer look like four. The tools help. The discipline is mostly in the preparation before the writing starts, not the writing itself.

What does your voice-switching process look like? Reply and let me know. This is one of those problems where everyone’s solution is slightly different and I always learn something useful from the variations.

In the course of our initial onboarding call, I learned of this situation as a cautionary tale. I saw it as an example.

Here’s the thing. There is a wide gap between “theoretically acceptable” and “actually good.” Most writers get replaced in the gray area between those two. Clients may not be able to articulate why something is unacceptable; however, they typically know when something doesn’t sound right, and they seek writers who can fix the issues before they blow up.

Over roughly two years of full-time freelancing, I developed my pre-delivery checklist. While each item can take anywhere from 30 seconds to several minutes, each has identified something that could have ultimately cost me clients. The items on this checklist grew organically based upon my own experiences identifying what slips through when I’m running behind (which is more often than I care to admit) and what I wish I’d identified prior to sending the article to clients.

The AI Smell Test

The first three paragraphs of any piece I receive in my inbox go through my “AI Smell Test,” which involves having me read them out loud. No skimming. Literally read out loud. Which sounds completely absurd when I’m sitting alone in my apartment in Woodstock at 7 p.m., but it works.

Any sentence that causes me to stumble due to its rhythmic inconsistency is marked. Any two consecutive paragraphs that begin with the same transition are marked. Any paragraph(s) that appear to be written by someone who has read everything, but never actually spoken to another human being, are marked.

While none of this qualifies as scientific method, this is merely the result of having read enough raw AI-produced content to recognize the typical cadence pattern. Raw AI-written material reads perfectly well during skims, but sounds robotic the instant it comes out of your mouth. Human-written material exhibits rhythm and cadence variation -- short sentences following long ones; fragments where warranted; etc. Rhythm of someone who is actively thinking about the sentence while attempting to generate it vs. generating it.

If I am uncertain about whether an article meets this criterion, I export it to Walter Writes and perform the detection checks via Turnitin, GPTZero, Originality.ai, and Copyleaks within the editor itself. If everything shows green across all detectors, I proceed. If any show red, I fix those before submission.

The Opening Sentence

Approximately 6 months ago, I began testing articles by exporting the opening sentence into a new document and asking: if this is the only sentence I’ve ever read from this writer, would I continue reading the next one?

Only a handful of opening sentences have passed this test. This does not mean they’re poorly written -- it simply means they’re functioning as a warm-up sentence rather than an actual opening. They introduce context before arriving at the core idea/thesis of the article. Typically, only the second or third sentence of the original draft functions as a true opening sentence.

I delete opening sentences more frequently than any other singular action during the pre-delivery pass. If I can eliminate the first sentence and retain no loss of meaning or clarity, that sentence is gone. This may seem obvious. However, consistently applying this principle is significantly harder than it seems -- primarily because you know exactly what the piece is about. The reader doesn’t.

That’s precisely the problem with poorly constructed opening sentences. Writers see beyond them. Readers don’t get that far.

The Brand Voice Pass

Next, I open the client’s brand voice documentation adjacent to the draft and read three paragraphs simultaneously. I’m looking for language that belongs neither to them nor to the model used to create the draft.

A “warm-but-expert” brand doesn’t use language like “it is imperative.” An informal SaaS brand shouldn’t use formal SaaS transitional language. Honestly, most of what this catch accomplishes is identifying instances when either my own voice bled into theirs, or vice versa -- or when the model’s voice somehow blended in and became comfortable. At the moment it occurs, this check is often much more intuitive than it sounds, particularly if you’ve spent several hours writing and your own internal tone calibration has been altered such that you’re now only focused on finishing the job.

Additionally, I ensure that the draft did indeed accomplish what the client specifically requested (as opposed to what I believed they requested). These aren’t always equivalent. Eventually, clients realize this difference.

The Walter Run

Even pieces that I created largely by hand undergo a run-through in Walter Writes Humanizer prior to delivery.

Basic for a tone pass. Basic for many client blog posts. More advanced when the draft was predominantly composed of AI-produced content requiring additional structural revisions. The extent to which a client utilizes rewriting functionality matters more than it initially might suggest. Basic will not correct heavily AI-composed content. Advanced will over-correct content that is otherwise mostly human-created. Getting this level right is part of the function of this check.

I utilize Walter Writes to review even hand-composed pieces because I compose flat when I’m exhausted. End-of-the-month content; late-in-the-day drafts; content written on truly horrible mental health days. The structural rewrites performed by Walter help identify areas in my own writing where it becomes mechanical. This is less common than failing to detect AI-generated tells; however, it occurs often enough that I do not omit this step.

Following completion of the run-through of the article in Walter Writes, I again inspect the detection scores from Turnitin, GPTZero, Originality.ai, and Copyleaks within the editor itself. If everything displays green across all four detectors, I proceed. If any display red, I resolve those issues prior to submitting the piece.

The Word List

There are five phrases that I systematically search-and-replace regardless of client preference:

* “Use” replaces “utilize”
* “Delete” or “Rewrite” replace “seamless” (or similar terms)
* Replace “leverage” as a verb
* Remove “it’s important to note”; remove whatever you were going to note.
* Rewrite nearly every instance of “whether” since most times it appears in a sentence there exists a cleaner sentence variant once reviewed thoroughly.

Each individually represents minor offenses on their own. As a collective set, they represent an article with a thinly veiled layer of human editing applied atop generated content. A client who reviews the article and silently wonders if it was generated by AI tends to not be able to reference a specific sentence(s) that caused their suspicion regarding AI generation. However, they can sense the cumulative effect of these changes collectively.

Four minutes of systematic find-and-replace is worthwhile.

The Phone Test

Prior to submission, I manually export the final draft into an e-mail message addressed to myself and open it on my cell phone for review without prior knowledge of what lies ahead.

Poor formatting is immediately apparent on smaller screens. Paragraphs that appeared fine in Google Docs become walls of text. Sentences that seemed manageable at 13pt font become unmanageable at mobile resolution. When I begin skimming an article I’ve authored on my phone as a reviewer would as well.

Additionally, I pull just the H2 heading names and attempt to read them sequentially to determine if they form a coherent narrative arc. If they do not -- then there is likely structural damage that cannot be fixed by revising individual sentences. While not every time necessary, I have completely reorganized pieces at this stage after realizing that the headings provided little more than labels rather than forming an inherent argumentative flow.

This step requires approximately five minutes to complete. More than any other item on this checklist, it identifies layout and structural flaws.

Why a Checklist Matters More Than Better Editing Skills

On most days, I execute each of these steps unconsciously. The checklist serves as protection for those days -- the fatigued days; the distracted days; the days when I’m writing about something I find boringly tedious. Those are the days when things tend to slip.

The checklist is not meant to supplant skill; rather it serves as a substitution for mood-based performance variability. Skill remains constant whereas mood does not. Good processes serve as assurance that my worst performances are above a certain minimum bar -- regardless of how high or low my creative performance level is on any given day.

From three years as a full-time freelancer: process provides consistency to quality control while creativity ebbs and flows -- regardless of mood or fatigue level at any particular moment in time. For eighteen months now, I have utilized some iteration of this checklist to revise submissions prior to delivery. Prior to establishing this checklist, I experienced higher-than-average numbers of revision requests based solely on client perception that my submitted work failed to meet their expectations -- despite meeting mine. The establishment of this checklist eliminated those differences entirely on a daily basis.

You need something akin to this checklist established yourself -- begin with patterns you personally struggle with. Mine included recognizing AI rhythm; producing flat openings; and losing vocabulary continuity between successive drafts. Your will differ accordingly.

Here’s what I actually use, ranked by how much each tool contributes to retainer income, with the reasoning. No affiliate links. No sponsorship. Just what’s been on my dock for 18 months without rotation.

ChatGPT, paid plan. Used for first drafts, outline generation, research summaries, and brainstorming angles. The paid plan matters because the model quality is better and the rate limits don’t interrupt my workflow. Free tier is fine for casual use. For production work the upgrade pays for itself in week one.

Walter Writes. The humanizer in Walter Writes is step two of every piece I produce. The structural rewriting and the in-editor detector compress my workflow more than any other tool I’ve added in three years. I’ve tested QuillBot, Grammarly’s rewrite, and a couple of smaller humanizers. Walter is the one I kept. The rewrite strength settings (Simple, Standard, Enhanced) and the detector breakdown across Turnitin, GPTZero, Originality.ai, and Copyleaks are what make it the daily driver. This is the tool I’d quit a client over before quitting.

Google Docs. I draft and edit in Docs. I share drafts with clients in Docs. I keep my running notes in Docs. There’s nothing fancy here. Docs is reliable, ubiquitous, and free. I tested Notion as a writing surface for two months and went back to Docs. Notion is good for project management and bad for actual writing.

Notion. The project management side of my work runs in Notion. One database for active retainers. One for content pipeline. One for client briefs. One for invoices and revenue tracking. I’ve tried Trello, Asana, ClickUp, and a few others. Notion is the one that survived because the database flexibility lets me restructure as my business changes without migrating data.

A simple invoicing tool. I won’t name mine because the category is full of fine options and the choice doesn’t matter much. What matters is having one. I went 18 months early in my career manually creating invoices in Word and chasing late payments by hand. Switching to a tool that auto-sends invoices, sends reminders, and tracks payment status saved me the equivalent of one client per month in administrative time. This is the cheapest, highest-ROI tool I added.

A reliable VPN. I work remotely and travel occasionally. A VPN is non-negotiable for security and for accessing some client portals that geo-restrict. The brand isn’t important. Pick a reputable one and pay the annual fee.

Backup internet. I have my main fiber connection and a mobile data plan with a separate SIM in a portable hotspot. Cape Town’s power and internet reliability isn’t always perfect. The backup pays for itself the first time a deadline lands during an outage.

A noise-cancelling pair of headphones. Boring. Worth it. I work in shared spaces sometimes and the difference between writing in a coffee shop with full noise cancellation and writing in the same coffee shop without it is roughly two hours of productive output per day.

That’s the core stack. Eight tools. Most of them I’ve been using for over a year. None of them are exciting.

What I tried and stopped using

A short list of things I tested and dropped.

Otter for meeting transcripts. Useful but I have so few meetings that the subscription wasn’t worth it. I now use ChatGPT to summarize the rare call I record.

Grammarly Premium. I have the free version. The premium tier was good but redundant once Walter handled the humanization layer and Docs handled basic grammar.

Several “AI writing assistants” that promised to be the entire stack in one tool. None of them did any one job better than the dedicated tools. The all-in-one pitch is appealing and almost always worse in practice.

A shared inbox tool for client communication. I tried it for a month. Killed it. My client volume doesn’t justify the overhead. Email and a labels system in Gmail handles everything.

A fancy time tracker. I tried three. None of them changed my behavior. I now just note start and stop times in Notion if I need to track a project’s hours, which is rare.

The pattern in what I dropped: tools that either solved a problem I didn’t really have, or tools that overlapped with something I was already using. The wins came from tools that did one thing well and integrated cleanly into the workflow that already existed.

What new writers should buy first

If you’re starting out, the order I’d buy in:

Reliable internet. Not optional. Whatever this costs in your market, pay it. Lost work hours from connection issues will cost more than the difference between basic and premium internet plans.

Paid ChatGPT or equivalent. Frees you from rate limits and gives you better drafts. Cheap relative to what it produces.

A humanizer. I’ve made my case for Walter. Try a free tier first. Commit when the math is obvious.

Notion or equivalent for project tracking. Free tier is enough for most freelancers.

Invoicing software. Don’t try to do it manually. You will lose money.

Headphones, if you don’t already have them. Sounds obvious. Most freelancers don’t have decent ones until they’ve been working for a year. They should buy them on day one.

Skip everything else until you have a specific problem to solve.

What I refuse to buy

Productivity courses. I’ve seen too many people pay $500 for a course that teaches what could be learned by following two solid blogs for a month. Free content is plentiful. Paid content is rarely worth it for working writers.

Conferences. Niche question, but worth saying. I’ve attended two and got more value from one paid coffee with a peer than from either conference. The travel time is a real cost. The networking is real but compressed and superficial. Skip unless your specific goal aligns.

Multi-thousand-dollar mastermind groups. The price-to-value ratio is almost always bad. Most of what gets shared is available in cheaper or free formats. The exclusivity is the product, and exclusivity isn’t a skill input.

Anything sold with the word “transformative” in the marketing copy. Honestly, this is just a heuristic, but it’s a reliable one.

To be fair, this is what works for me. Other writers thrive on different stacks. The point isn’t the specific list. The point is being honest about which tools are doing real work and which ones are decoration. Most writers I know underestimate how much friction comes from tools that aren’t quite earning their place.

The boring stack wins because it stays out of your way. Every tool that disappears into your workflow is a tool that adds compounding return. Every tool that demands your attention is a tax. Audit yours this week. Cancel three things. Notice what doesn’t change.

This is the problem. Freelance writers view the fact that clients are switching to using ChatGPT (or similar) as a sign they need to decrease their prices. The numbers seem simple. Since AI is inexpensive, why shouldn’t writers be inexpensive too? But this completely misrepresents the whole issue.

The clients who switched to using ChatGPT were never your clients. Those clients were transactional. They were interested in obtaining any words possible. When those words became free, they exited. There’s little chance those clients will ever pay you more than they did initially. There’s little chance they’ll ever pay you the same as they did prior to AI. And now there’s virtually no chance they’ll pay you less. Most importantly, they’ll pay you nothing. Lowering your price to attempt to recapture these clients is essentially pricing yourself lower in an effort to pursue ghosts.

The clients who truly paid you were paying you for something AI cannot provide. Strategy. Voice. Judgment. A persuasive argument for your intended audience. Those clients remained when AI became cheaper, because what they purchased was not just typing. Cutting your price will indicate to those clients you don’t understand what you provided was valuable. Many lose faith in you. Others quietly move you down on their priority lists. Those clients remaining will use the new lower price as an anchor during future negotiations.

It’s very simple. Low priced offerings do not compete against AI. They compete against low priced. The only people you may attract with low priced offers are individuals who will offer to drop their prices again next quarter and the quarter after that. That is a race to zero, and AI wins races to zero by definition.

How to compete above the typing layer

I’ve maintained my rates unchanged for 14 months and increased them once in that period. My decision to maintain rates was not based upon my status as a unique individual. It’s based upon the fact that I’ve developed three things into how I sell.

1. A workflow I can explain within a single breath. AI draft, humanization via Walter Writes, manual edit. Three steps. Twenty minutes to demonstrate on a call. The workflow makes me faster than writers who produce drafts manually and more reliable than writers who submit raw AI drafts. Clients tend to rely on processes rather than talent statements, and the process can be verified.

2. Rules regarding what I will and will not accept. I won’t accept any engagement until I receive a brief. I won’t accept projects shorter than 600 words. I won’t accept clients who bargain on our initial conversation. These rules are not related to being choosy. They’re about costs. Inefficient projects consume time, and time is the true budget. Declining opportunities quickly allows me to say yes to the right ones. The clients attempting to bargain you down to $.05/word would likely provide poorly defined briefs and request revisions four times. Either way, the opportunity for profit disappears.

3. I provide something real prior to quoting. I don’t pitch using a presentation slide deck. Instead, I pitch by providing a complimentary 200-word draft of a project they require. The draft includes the actual brief I’d develop, the approach I’d apply, and a sample of the tone used in the piece. The process takes approximately thirty minutes. More often than not, pitching using a complimentary draft closes more deals than linking potential clients to my portfolio.

The rate question, specifically

Many of the writers I’m familiar with charge between $0.08 and $0.30 a word. Any rate below the lowest end of this spectrum is impossible to sustainably support if a writer wishes to generate sufficient income from their writing efforts. The mid-range supports transactional writing but doesn’t provide adequate funding for long-term retainers. The highest end generates sufficient income to operate a small business if the writer can secure retainers with three or four clients.

The best method for determining whether you are properly charging is by testing your current rate against your desired level of income and hours worked annually. If your desired income can’t be generated with your current rate while maintaining a reasonable number of hours worked each year, then the rate needs adjustment, not your strategy. Lowering your rate will not correct your financial equation. Increasing it could, if you can justify the higher rate.

Justifying a higher rate

You can’t justify a higher rate based solely upon your background or credentials alone. You must focus on demonstrating how your writing will benefit the client. “I charge $X due to my twelve years of writing experience” is an ineffective justification. “I charge $X because my previous three retained clients averaged an organic traffic growth of 28% over 90 days” is a much stronger one. The first statement focuses on you. The second focuses on results. Clients purchase results, not biographies.

Not every writer has access to traffic data showing how their writing performed. If you don’t possess such data, create it. Begin tracking the next three articles you write. Document the keywords, publish date, and position ranking at 30, 60, and 90 days post-publication for each article. As stated earlier, AI has made content inexpensive. Creating a correlation between your writing efforts and measurable results that can be validated by your client represents the best defense available today.

Volume?

Several writers discuss their ability to make the math work by producing additional articles faster, even at reduced rates. While theoretically viable, this concept fails in practice. Producing large volumes of articles at low rates means more clients, more briefs, more invoices, additional rounds of revisions, and additional administrative tasks, resulting in your hourly rate decreasing more rapidly than your production increases.

As stated earlier, the three-step workflow is not designed to allow writers to produce three times as many articles at the same rate. Rather, it’s designed to enable writers to produce the same number of articles in half the time, then either take on additional retainers or reclaim their time for strategic planning. Writing and editing cycles remain separated. That’s actually where low-priced volume work falls short.

In recent history, I’ve witnessed two friends attempt this model. Both have since ceased operations. Not due to the difficulty involved in performing writing assignments, but rather due to the frustration experienced by both writers as they attempted to perform at high volumes at low rates, and subsequently felt defeated as every interaction with clients seemed like another defeat. Ultimately, both sets of clients viewed writing as an inexpensive commodity, and consequently valued neither the writers nor their work.

What to do this week

If you believe you’re currently undercharging, increase your next quoted rate by 25%. Not with existing clients, but with any new leads entering the pipeline. Observe how they react. If they decline, there was no way you would have generated revenue from them anyway. If they agree, you gain new data.

If you’re experiencing losses due to ChatGPT, audit your losses. Did those clients pay you for an upper-level service (strategy and voice)? Or did they merely pay you for typing? Regardless of which it is, if they paid you merely for typing, they were always going to depart regardless of what technology was employed. If they paid you for an upper-level service, determine what caused them to cease purchasing those services from you. Frequently, it’s not the emergence of AI itself, but instead changes in budgets, internal hires, or shifts in strategies that drive this behavior. Don’t automatically conclude that ChatGPT was responsible for their departure. Ask.

If you don’t have a workflow capable of being demonstrated, create one over the weekend. AI draft, humanizer, manual edit. Three steps, three tools. Then go close better clients with it.

Honesty time: the writers succeeding today are not necessarily offering services at the lowest price levels. They’re selling more defensible services. Their workflows, refusal criteria, and complimentary draft pitches are their defenses.

Low-cost services represent an admission of defeat.

The expensive part of content writing is not the writing. It’s the misalignment. And misalignment almost always traces back to the brief.

A good brief is a product. It contains a problem statement, an audience, a desired outcome, three reference pieces, the unique angle the brand is going to claim, and a list of what the post is not about. Five minutes of clarity from the client compresses two days of work into four hours. This isn’t theory. I’ve measured it on my own retainers.

A bad brief is a sentence. “Write us a post about AI in marketing.” Or worse, no brief at all, just a Slack message saying “let’s get a post up this week.” Bad briefs don’t save the client time. They transfer the work to me, and I either guess or ask questions, which means more meetings, more revisions, more cost.

Here’s the math. A vague brief produces a draft that’s directionally wrong. The client reads it, can’t say what’s wrong, but knows it’s not right. They send vague feedback. I rewrite. Repeat twice. Eventually we converge on something they accept, often because they’ve run out of patience, not because the piece is actually right. Total time: 12-14 hours. A good brief produces a draft that’s 80% right on the first pass. Light revision lands the rest. Total time: 4-6 hours. The output quality is higher. The relationship is healthier. Everyone is paid the same.

So why don’t more clients write good briefs? Because brief writing is harder than people assume. Articulating what you want forces you to actually decide what you want. Most founders don’t have time to make those decisions every week, and content lives in the queue of “we’ll figure it out when we see it.”

Here’s the thing. The fix isn’t to refuse to work without a perfect brief. The fix is to write the brief with the client in the first 15 minutes of the engagement, then reuse the structure for every subsequent piece. I have a one-page template I use with every retainer. Five fields. The client fills in three of them. I fill in two. We never start a piece without it.

The fields:

Working title. The client picks this even if it changes later. Specific is better than clever. “How startups should think about AI content costs in 2026” is fine. “AI and the future of marketing” is not.

Audience and stage. Who’s reading this and where are they in the buying process. “Founders evaluating content vendors” reads differently than “marketing leaders looking to defend a content budget.” Same topic. Different post.

Desired outcome. What’s the one thing you want a reader to do or believe after reading this? Not three things. One. If the client lists three, I push back until they pick one.

Three reference pieces. Two examples of what good looks like to the client, one example of what they don’t want. The contrast tells me more than any positive example alone.

The angle and the not-about. The argument the post will make and the territory it deliberately won’t cover. The not-about list is the part most briefs skip and it’s the part that prevents scope creep.

Once the brief is solid, the rest of my workflow runs predictably. I draft with ChatGPT against the brief, run it through Walter Writes to handle the structural rewrite, then edit for voice and specifics. Without a tight brief, the AI draft step is useless because I’m asking the model to make decisions that should have been made by the human paying for the work. The humanizer can’t fix a directionally wrong draft. Walter’s adjustable rewrite strength is great for tone calibration but no humanizer can recover a piece that’s about the wrong thing.

A side note that matters. Vague briefs don’t only damage individual projects. They damage retainer pricing. When I started, I used to absorb the cost of vague briefs and rewrite drafts at no charge. After two retainers ended unprofitably, I changed my contract. Brief revisions are billed. Major scope changes mid-draft are billed. The client’s option is to either invest in the brief or pay for the consequences of skipping it. Most pick the former once it’s their money.

I’ve seen writers try to solve this with elaborate intake forms and 40-question questionnaires. Those don’t work because nobody fills them out. The five-field template works because it’s small enough to write in one sitting and clear enough to enforce. I’ve also seen writers try to skip the brief entirely and “build trust through volume.” That’s a trap. You produce more wrong work faster, the client gets frustrated, and you both blame each other.

Some practical patterns I’ve learned.

The brief should be written, not verbal. Verbal briefs change between Tuesday and Friday. Written ones hold.

The reference piece the client doesn’t want is more useful than the two they do want. It tells you what to actively avoid.

If a client refuses to fill out the brief or sends three sentences, that’s data. Either they’re not invested or the project isn’t ready. I’ve started declining engagements that won’t commit to a brief. The opportunity cost of bad-brief work is too high.

If you’re a writer reading this, write your own template tonight. One page. Five fields. Use it on your next intake call. Watch how the conversation changes when the client has to actually decide what they want.

If you’re a founder reading this, your content writer is not the bottleneck. The brief is. Pay your writer to write it with you the first time. After that, fill in the fields yourself in 10 minutes per piece. The output will sharpen by the second post and lock in by the fourth.

To be fair, sometimes the brief is solid and the post still misses. That’s normal. But when the post misses with a solid brief, the conversation is productive: we know what was off, we know how to fix it, the next iteration converges fast. When the brief is bad, every conversation feels personal because nobody knows what good looks like. That’s the unwinnable mode.

The reality is, the more you type in conjunction with ChatGPT, the more the rhythm of how you write with the tool will begin to show itself in the way you write when the tool is closed. Your articles will begin to sound like this:

You’ll find yourself opening pieces with rhetorical questions that you’d never thought to pose. You’ll find yourself saying things like “leveraging,” in your sentences. You’ll find yourself qualifying each statement with “in some cases” or “depending on context,” simply because the model does so. And, eventually, you won’t sound anything like yourself.

That’s important, since the whole reason a client hires you instead of typing something themselves (and then asking ChatGPT for assistance), is because there are differences in tone between you and ChatGPT. When those differences disappear, your ability to compete with a model that can write similar quality material at a fraction of the cost disappears.

The five default phrases I remove from all my writing:

1. Over-formal openings: “In today’s rapidly changing digital environment.” “As companies continue to adopt… “ “With the increasing reliance on… “ Remove the entire opening paragraph. Write about what you’re really trying to get across. If what you’re really trying to get across can’t be opened with, then the angle needs to change.

2. The morality clause: “Although AI brings many advantages, we must consider the potential ethical implications.” It’s nothing more than a tax that AI must pay on virtually every topic where ethics may possibly apply. Writers either address ethics head-on, or ignore it entirely. Writers shouldn’t hedge their bets.

3. The ultimate wrap-up: ChatGPT loves “Ultimately, success comes down to… “ So do any articles that conclude with a clean moral. Real essays end hard. They end somewhere. The ultimate wrap-up indicates the writer didn’t know how to end.

4. Empty intensifiers: “Truly transformational.” “Extremely impactful.” “Very critical.” None of these adjectives provide additional information. They merely add length. Remove them and re-read the sentence. It usually reads much better after removing them.

5. Vague specifics: “Numerous studies have demonstrated… “ “An ever-growing amount of research… “ “Most industry professionals… “ These phrases attempt to reference sources without referencing any particular source. If you have a study to reference, identify it. If not, say “I believe.” Either method is far more credible than pretending to be authoritative.

I run my drafts through Walter Writes as much for its structural disruption of these default patterns as for anything else. The rewrite strength options allow me to move beyond mere paraphrasing toward actual restructuring of sentences. But the humanizer does approximately 70% of the work. The other 30% is me, manually finding each of the five default phrases I’ve outlined above and rewriting them myself.

Why default phrases exist

ChatGPT’s default voice is consensus voice. The model was trained on an enormous amount of data and evolved into the safest, most generic way possible of writing about virtually any topic. That’s why every piece of writing generated by raw AI output appears somewhat like a corporate blog from 2018. All of corporate America’s blogs from 2018 make up the majority of ChatGPT’s training data.

Your voice, non-consensus

Non-consensus is unique to your voice. Your voice contains opinions based upon your experiences, stories based upon your experiences, and an opinionated cadence formed by what you’ve read, where you grew up, and whom you write for. This non-consensus nature is what generates interest in your content. And this is what sets you apart as a writer from others.

Every time you accept an AI-generated phrase you would have never typed yourself as part of your writing process, you’re accepting consensus rather than non-consensus. Accept this often enough and your voice will converge with that of the model. You’ll be an expensive means of generating mediocre AI content.

The solution is to edit: not for grammar, but for personality

After Walter restructures the rhythm of your article, I review your article and ask myself two questions:

Would I say this in conversation?

If not, rewrite.

Would I open an article with an opening sentence such as this if I wasn’t being paid?

If not, rewrite the opening sentences.

Does my close sound like me or does it sound like a TED Talk?

If TED Talk, eliminate it.

At first this will take time. Eventually it gets fast because you’ll begin spotting defaults instantly. Eventually it stays fast permanently because the AI tics won’t show up in your drafts as often. You’ll have learned how to avoid them during prompting.

A practical exercise

Open a 600-word post you created within the last month. Determine how many times each of these phrases appears in your post: leverage, utilize, streamline, cutting-edge, game-changer, revolutionize, seamless, robust, ultimately, in conclusion, in today’s. Regardless of how many occur, any number greater than zero indicates that the model is beginning to bleed into your voice. Any number greater than three and you’re creating ChatGPT content under your byline.

Read your post aloud to someone who writes for a living. Observe how they react. Wherever their focus shifts away from your voice and onto something else, circle that sentence. Whenever they exclaim “Huh?” whenever they say that sentence outloud, that sentence is yours. Every sentence where they drifted off is a product of the model.

To be clear, AI’s default voice is not terrible writing. It’s competent writing, and competent writing is currently considered acceptable minimum standards in our current market. The issue here is that competent writing is no longer a ceiling. It’s a floor. If your writing sits at the floor, you’re competing with a free tool that produces content of similar quality at a fraction of the cost.

There are several writers I admire who are attempting to break down their AI usage. They’re drafting more of their articles by hand and using AI primarily for research purposes and outlining purposes. Other writers are still drafting more heavily with AI, but they spend double the time editing as well. Ultimately there’s no correct answer. There’s merely the fundamental principle that any article produced should sound like a person, and that person should be you, not the model. I’m evaluating Walter’s output against QuillBot’s. The structure-based rewriting holds up slightly better for my specific use case, but it’s clearly the editing layer which determines whether my voice remains intact.

Those writers who remain relevant are going to be those writers who are able to create content using AI without appearing to be AI-created. Those writers will replace anyone who can’t maintain this self-discipline.

Which default phrase have you recently found yourself absorbing from an AI model? Tell me via reply, and I’ll include the best ones in a future post.

Clients don’t pay me to type. They never paid me to type. Clients pay me for the space between a brief and published work. That space is taste, judgment, and accountability. AI has essentially made typing free. It’s done nothing to the layer above.

What Clients Are Paying Me For

When a client pays me to write a 1,200-word blog post, here’s what’s really on their invoice.

They’re getting the first round of strategic thinking, which happens before any AI tool is involved. I’m reading their last six published pieces, their pricing page, their two highest-traffic competitors, and figuring out what argument they can plausibly own. AI can give options. It can’t decide which option fits their brand position.

They’re getting the edit, which is where my margin lives. I use ChatGPT to draft, run it through Walter Writes for the structural rewrite, then manually edit for specifics. The edit usually takes up around 40% of the time on a piece. It’s where I add the example that grounds the claim, the personal aside that breaks the rhythm, the contrarian sentence that earns the link. Honestly, it’s the part nobody sees, and the part that decides whether a piece performs.

They’re getting consistency of brand voice. Across 12 posts in a quarter, if you want them to sound like the same person wrote all of them, you need human intervention. Even using the same prompt twice, ChatGPT drifts. It picks up new habits over sessions. My mental model of how their brand sounds helps with editing toward consistency. Founders underestimate how important this is until they read three of their own posts back-to-back and notice one doesn’t fit.

They’re getting the no. Most of my value comes from the briefs I push back on. The angle that won’t land. The keyword that’s too competitive. The post that reads defensively instead of confidently. AI will write whatever you ask it to write, including the bad version. I won’t. That’s the difference between a contractor and a vendor.

They’re getting long-term relationship context. After 18 months on retainer with one client, I know their past three product launches, the analyst they want to impress, and the customer quote they keep forgetting to use. That context produces better writing. It also makes me harder to replace once established. ChatGPT starts every session blank. I don’t.

They’re getting fact-checking, which has become more relevant in the last year than it was three years ago. AI confidently generates wrong stats, fake quotes, and outdated comparisons. Walter Writes doesn’t fact-check, no humanizer does, and that’s not a flaw, that’s the scope. The fact-checking is on me. I cross-reference each number, link each claim, and cut anything I can’t verify. Clients trust the writing because they trust the accuracy behind it.

Things That Have Changed and Things That Haven’t

Three things have changed.

1. Speed expectation. Clients who used to wait three days for a draft now expect two. AI shortened the production timeline and the market repriced. I matched the new pace.

2. Volume. With the three-step workflow, I produce more in a week than I used to in two. That means I can take retainer commitments that were previously outside my range. Retainers are stickier and better paid than one-off projects, so this matters for income stability.

3. The conversation about AI is finally over with most clients. A year ago I had to explain that yes, I use AI, no, that doesn’t mean the post is generic, here’s the workflow, here’s why the rate is what it is. Now most clients ask what tools I use and we move on. A few still treat AI as a moral question. Those usually aren’t a good fit anyway.

What remains unchanged is what content has to do. Build trust, signal expertise, rank for the right queries, support the sales motion. AI didn’t change the goals. It just made it cheaper to produce one input to those goals.

If You Write for a Living

Don’t try to compete on price with raw AI output. You’ll lose. Buyers who care only about the cheapest possible content were never going to pay your rate anyway. Compete on the layer above: strategy, voice, judgment, and accountability.

Develop a process you can stand by. Mine is AI draft, humanization through Walter Writes, and manual edits. I can describe it in one sentence and demonstrate it in 20 minutes. Clients trust process more than talent claims, partly because process is verifiable.

Invest in tools that compress your time. I’ve tried QuillBot, Grammarly’s rewrite function, and several smaller humanizers. Walter Writes stood out because the structure-level rewrites hold up across the variety of brands I write for. The detector being in the same editor is what makes the time savings real, not theoretical.

Price for the layer above the typing. Itemize it if you have to. “Strategy and brief” is a line item. “Editing and brand voice” is a line item. “Drafting” is a line item. Some clients will object. The right ones won’t.

This year will separate writers who built a workflow from writers who didn’t. The retainer money is in the workflow. The freelance scramble is in the lack of one.

I told him the real giveaway didn’t have much to do with vocabulary or subject matter. It was the flow.

That’s the issue no one really explains well enough. Raw AI output isn’t terrible because of word choice. It’s terrible because all the sentences are approximately the same length. All the paragraphs look like they’re cut from the same template. All the transitions hit exactly the same beats. The text is grammatically correct and utterly emotionless, like a robot wearing a suit. You can tell something’s off long before you can figure out what specifically is wrong.

After editing two years of raw AI drafts, I’ve found four things that indicate AI was used.

1. Sentence length uniformity. ChatGPT writes in consistently medium-length sentences. One comma per sentence. No fragments. No complex sentences with multiple sub-clauses. Real writing has rhythm. A short sentence. Followed by a longer sentence that loops back, qualifies, and stretches the idea further than feels efficient. Short again. A human can do that. An AI can’t. You have to tell it to.

2. Connectors. Watch for “moreover,” “furthermore,” “in addition,” “however,” “that said,” “ultimately,” and “in conclusion.” Those are AI’s favorite way of making sure you understand how everything fits together. Humans don’t rely on them that often. We use colons. We leave a whole paragraph to breathe. We trust the reader to follow along without holding their hand.

3. Wraps. AI likes to add a neat little wrap-up at the end of a paragraph summarizing what was just discussed. There’s no reason for this unless you want to pad the word count and reduce the trust between yourself and your audience. A human writer would’ve left it alone.

4. Hedged specifics. Instead of pointing to one study, AI says “many studies suggest” something. Instead of describing a specific scenario, it says “in some cases.” Real writers take sides. Or they admit they don’t know. AI tries to do both at once.

Since I’m editing my drafts through Walter Writes, and it makes structural edits based on structure rather than style, it naturally addresses all of these areas. It varies sentence length, disrupts predictable transitions, and introduces what the company calls burstiness, the technical term for what I’ve been describing as rhythm. When I’m particularly unhappy with a draft, I can increase the rewrite strength. I tested QuillBot for similar structural edits, and the output read like the same robot in a slightly different suit. Rhythmic issues aren’t fixed with synonym swaps. They’re fixed with restructuring.

There’s also a detection step worth mentioning. After humanizing my content, I want to see the AI likelihood score before sending anything to a client. That’s why I stayed with Walter Writes. I can humanize, score, and adjust within the same editor without opening another tab or subscribing to another service. Originality.ai works fine as a standalone detector, but I didn’t want two separate services.

Walter Writes humanizer gives you significantly better raw material than an unedited AI draft. But it doesn’t give you a finished piece. Even after heavy humanization, you’ll likely need further revisions before you’re satisfied with the quality. The tool improves the raw material. The rest is still on you.

A metaphor: AI gives you a basic furniture frame. The humanizer sands and stains the wood. You still have to upholster it.

Practical Exercise 1: Find an existing AI draft. Read it aloud. Wherever your voice drops off, mark that sentence. Wherever you get stuck, mark it. Wherever you feel like skimming, mark it. Those marks are the work.

Practical Exercise 2: Pick two writers in your industry whose work you enjoy. Compare their sentence lengths in a single paragraph. Count the syllables in their first three sentences. Then count the syllables in three randomly generated AI sentences. The difference becomes obvious once you start measuring.

I’ve written before about my three-step process: AI draft, humanize, manual edit. The order works because each step solves a different problem. The draft handles the blank-page problem. The humanizer handles the rhythm and structure problem. The manual edit handles the meaning, voice, and specifics problem. None of those steps can do the others’ jobs.

A note on burstiness

Burstiness isn’t jargon. It refers to variability in sentence length and complexity. Human writing expands and contracts. AI defaults do the opposite: consistently sized sentences, consistently sized paragraphs, consistently sized cadence. Detectors flag this directly. So do readers, even when they can’t name what they’re noticing.

If you’re making money by writing and not paying attention to rhythm, you’re missing your greatest opportunity for professional differentiation. Vocabulary is easy. Grammar is easy. AI handles both well enough. Rhythm is where humans still beat the machines, and it’s likely where clients are paying for your skills, even if they don’t know how to ask for them by name.

Three things to try this week

1. Read your latest published piece aloud and mark every flat sentence.

2. Next time you generate a draft with AI, set your rewrite level higher than feels necessary.

3. Most writers under-edit. Good writers lightly over-edit and then trim back.

Many writers I speak to view using AI as either a magic trick or an ethical dilemma. It’s neither. AI is simply an automated drafting tool that produces flat, emotionless first drafts. What happens after the first draft is what makes the difference. Honestly, the writers who understand this first will likely be the ones to maintain their retainers.

Here is the workflow that pays my bills.

Step 1: The AI draft

I open ChatGPT. I enter all the previously approved article examples provided by the client along with any client-approved brand voice notes. I request an outline first. Once the outline has been generated, I request a draft based on that outline. I don’t ever accept the first draft. I push back on tone, structure, or angle until satisfied. At the end of Step 1, I have a very rough but directional first draft.

There are several things many people do incorrectly when creating their own workflow using AI. Some writers don’t provide adequate prompting for the AI, then complain about receiving generic fluff. Others accept a single draft and call it good. Either approach creates suboptimal results. The draft at the end of Step 1 should be 60% of the final word count and 100% of the structural framework. Nothing more. Nothing less.

To be honest, I’m not going to pretend ChatGPT is perfect. While it certainly generates impressive output, there are areas that need improvement. Specifically, ChatGPT tends to hallucinate, it loves using em-dashes, and it generates the same five transition words across every industry. None of this matters at this stage because I’m only concerned with obtaining raw material for Step 2.

Step 2: Humanizing the first draft

At this point I run everything through WalterWrites’ AI Humanizer. Before committing to WalterWrites, I tested four serious humanizers. The primary reason I stayed with it over the other options was its ability to structurally rephrase how ideas are presented within the original draft. Raw AI output requires this type of rephrasing. Simply substituting synonyms maintains the AI-based rhythm, and rhythm is typically a dead giveaway that something was written by an algorithm.

Another primary benefit of WalterWrites is the integrated detector. After humanizing the text, I can see on the same screen an AI-likelihood score against Turnitin, GPTZero, Originality.ai, and Copyleaks. If any of these tools flag the content, I increase the rewrite strength and run it again. I’m not copy-pasting between two different tools. I’m not paying for two separate subscriptions. The entire step generally takes me under a minute.

I’ve used QuillBot in the past and Grammarly’s rewrite feature as well. Both are suitable for sentence-level polishing. Neither performs structural pattern shifting. If your clients require content that reads like a person wrote it, and most of mine do, significant work has to happen in Step 2.

Step 3: Manual editing

This part of the workflow has been largely ignored by nearly everyone in the space. It’s also where my expertise lives. I open the humanized content in Google Docs and manually revise each line.

I remove all em-dashes. AI loves them, and humanizers rarely eliminate them completely. I replace them with hyphens, colons, or I rewrite the sentence entirely. Every instance of “whether” gets cut. I search for repeated transitions like “moreover,” “furthermore,” and “in addition,” then remove or replace them. I also verify that the opening reads like the client’s brand rather than a generic blog opener.

After those edits, I read all the revised content aloud. If a sentence sounds awkward, I fix it. If a paragraph can be cut, I cut it. If a section needs a personal anecdote or a specific data point to land, I add it.

This is where my true expertise resides. AI produced the raw clay. The humanizer made it easier to shape. The manual edit is the actual sculpture.

Why this matters

A few months before posting this, one of my clients tried to negotiate my rate down. Their argument: “We could create this content using ChatGPT for free.” I sent them the humanized, edited version of one of their published articles alongside the raw ChatGPT draft I started from. They didn’t bring it up again.

Clients don’t hire you to use or avoid AI. They hire you for taste, judgment, and a final product that sounds like them. The three-step workflow lets me produce greater volume without compromising quality, which means I can take on retainer work without burning out. That’s the whole game.

The adjustable rewrite strength in WalterWrites is more valuable than I initially expected. Most of my content uses Standard. For clients with strict compliance tones or intentionally dry voices, I drop it to Simple and edit more aggressively by hand. For Substack-style posts and creator newsletters, I run Enhanced and barely touch them after. Three levels is enough. More would slow me down.

Practical notes:

• Don’t paste client-sensitive content into any tool you haven’t reviewed for privacy policies. WalterWrites works for my use case, but review their terms relative to yours.

• Always read your completed piece end to end before sending it to the client. AI can produce incorrect facts confidently, and humanizers don’t fact-check.

The writers who’ll thrive in the next two years aren’t the ones who refuse to use AI, and they aren’t the ones who paste raw output into client briefs. They’re the ones who built a workflow they trust and learned to edit better than the tools do.

What’s your current AI workflow? I’m curious whether others are building structured processes or still doing it ad hoc. Reply below.

This is what I use today, ranked in terms of how each tool impacts my retainer income, along with the reasoning. No affiliate links. No sponsorships. Just what has sat at my dock for 18 months without being rotated.

ChatGPT, paid plan. I used ChatGPT to write the initial drafts of articles. Also used it to create outlines, provide summary versions of the research I conducted, and develop ideas for possible angles. The reason the paid plan is necessary is that the quality of the models is greater and there are no rate limits to interfere with my workflow. The free tier is suitable for occasional users. The paid tier of ChatGPT pays for itself in week one in terms of production work.

Walter Writes. Walter Writes is the humanizer that I run second in my process. The structural rewrite capabilities and in-editor detector of Walter Writes reduce my workload more than any other tool I’ve incorporated into my workflow in the past three years. I’ve tried Quillbot, Grammarly’s rewrite option, and a couple of lesser known humanizers. Walter Writes is the tool I’d walk away from a client contract over before walking away from Walter Writes.

Google Docs. I draft and edit in Google Docs. I send copies of drafts to clients via Docs. I maintain my running notes in Docs. There’s little to report here. Docs is reliable, everywhere, and free. I experimented with using Notion as a writing platform for two months and returned to using Docs. While Notion is a great platform for managing projects, it’s a poor choice for actual writing.

Notion. The project management aspect of my business operates within Notion. One database for current clients. One for content pipeline. One for client briefs. One for tracking invoices and revenue. I’ve attempted several platforms including Trello, Asana, ClickUp, etc. Notion remains due to its ability to allow me to reorganize databases as my business evolves without losing data migration.

A simple invoicing program. I won’t identify mine since the choices are numerous in this space and ultimately none are significantly different. Having an invoicing solution is what’s critical. I spent 18 months at the beginning of my career generating invoices in Microsoft Word and trying to find clients late on their payments. Once I began using an automated invoicing program that automatically sent invoices, reminders for overdue payments, and tracked payments received, I reduced my administrative time equivalent to retaining one additional client per month. This is the least expensive, highest ROI addition to my “stack.”

A reliable VPN. I operate outside of my primary location as a remote worker and may be traveling periodically. A VPN is essential both from a security perspective and for accessing client portals restricted based on geographical area. The brand is irrelevant. Simply select a reputable VPN and pay the annual fee.

An alternate internet service provider. In addition to my main fiber service provider, I also have a mobile data plan with a secondary SIM card installed in a portable hotspot device. Cape Town’s electrical grid and internet connectivity can’t always be relied upon. The alternate internet source pays for itself the first time an outage occurs while a deadline is looming.

Noise-canceling headphones. Unexciting. Worth it. When I choose to work in public spaces and need to minimize distractions caused by ambient noises, having noise-canceling headphones increases my output substantially over what it would have been without such a feature: approximately two hours more per day.

These represent my top eight tools. Most of these tools I’ve been using for over a year. None of them are exciting.

Tools That Failed Me

A small collection of items I evaluated and decided against implementing.

Otter for meeting transcripts. It worked fine but since I typically only meet once per month, the cost of maintaining a paid subscription wasn’t justified. Today, if I need to capture a rare recorded phone call with a client, I simply have ChatGPT summarize it.

Grammarly Premium. I use the free version of Grammarly. While Grammarly Premium was useful, after Walter Writes became part of my process and Google Docs took care of basic grammatical errors, the premium features provided minimal benefit.

A handful of AI-based writing assistant programs that claimed to replace my existing writing workflow. None of them performed any single function better than any of the individual tools that make up my workflow today. Anytime someone pitches an all-in-one solution for anything related to writing it seems appealing at first, yet nearly always results in poorer performance than taking advantage of multiple tools individually designed for particular aspects of writing workflows.

A shared inbox application for client communications. After testing it for one month, I canceled it. At present, the amount of client correspondence doesn’t support the additional expense associated with a shared inbox solution. Client email and organization systems in Gmail handle all communications effectively.

Time tracking applications with bells and whistles. Although I tested three time tracking solutions, none impacted my usage patterns in a positive manner. If I need to document time spent on a project, I simply log start and end times directly into Notion, which is rare.

The common theme among the tools that I abandoned: tools that addressed problems I never had, or tools that duplicated functionality I was already receiving from another tool. Conversely, the success stories were from tools that focused on one task very well and fit seamlessly into an existing workflow.

Which Tools Should New Writers Invest in First?

When starting out, the order that I recommend investing in:

1. Reliable internet connection. Not optional. Whatever it costs in your local market, spend it. Time lost due to connection loss will far exceed the cost difference between standard and best internet plans.

2. Paid ChatGPT or equivalent. Allows you to avoid rate limits and provides better drafts. Very affordable compared to what it generates.

3. Humanizer. I’ve made the case for why Walter Writes is best. Test a free version first. Commit when it becomes obvious.

4. Notion or equivalent for project tracking. The free tier is sufficient for most freelance writers.

5. Invoicing solution. Do not attempt manual invoicing yourself. It’ll waste money in the long term.

6. Noise-canceling headphones. May seem obvious. Most freelancers wait until they’ve been working for at least a year before purchasing proper ones.

Wait on everything else until you’re solving a problem that requires it.

Which Purchases Do I Refuse to Make?

Productivity courses. I’ve seen too many people spend $500+ on courses teaching concepts that could be learned in one to two months reading two reliable blogs. Free resources are abundant, and paid resources are seldom worthwhile for producing writers.

Attending conferences. Niche comment, but worth mentioning. I attended two and received more value from one conversation with a peer over a $5 cup of coffee than from either conference. Travel time represents significant expense, and networking experiences tend to be shallow and condensed. Unless your specific objective aligns perfectly with attending conferences, skip them.

Membership and subscription fees over $1,000 per month. Price-to-value ratios are nearly always inferior in multi-thousand dollar mastermind groups. Much of what’s being exchanged among members exists freely or at lower-cost alternatives. Exclusivity is what they sell you. Exclusivity is not an input factor in your skills base.

Anything marketed as “transformative.” Honestly, this is just one heuristic, but it’s historically proven effective in identifying inferior products.

As stated previously, this represents what works for me personally as a writer. Other writers succeed with different sets of tools, products, and processes. The goal isn’t necessarily the unique set of tools themselves. It’s providing accurate information regarding which tools are adding true value versus those merely serving as eye candy.

The majority of writers I know underestimate the amount of friction created by tools that aren’t truly pulling their weight.

The “boring stack” succeeds because it doesn’t get in your way. Each tool that integrates into your workflow without causing distraction creates cumulative returns. Each tool that requires your attention creates a tax. Review your own “stack” this week. Remove three unnecessary tools. Note what doesn’t change.

What’s the one tool on your stack you can’t live without? Which one should you have removed a year ago? Respond via email reply.