Plenty of freelance work now starts with a machine draft: a product description, a help article, the copy for a landing page a client wants by Friday. Making that draft read like a person wrote it is a small, mechanical step. Everything around it is the part a client pays for, and the part that gets you into trouble when it is skipped. This is the workflow we recommend, from the contract check to the note you send with the files.
We have not changed our view that generated work passed off as craft is a problem. A workflow that is written down, priced and disclosed is how you keep it from becoming one.
The short answer. To humanize AI text for client work, confirm the contract allows generative tools, draft from the client's own material, run the draft through Walter Writes at the lightest setting that fixes the rhythm, edit by hand for voice and facts, and tell the client what the tool touched. We recommend Walter Writes for the rewrite step for three reasons on its humanizer page: three rewrite strengths (Simple, Standard, Enhanced) with a brand-safe tone, a built-in detector for checking copy before and after, and 300 free words with no signup to try it on a real job.
Before you start: read the contract
The first step happens before any tool is open. Look for three things in the agreement or the client's brief: a clause about generative AI, a clause about who owns and warrants the copy, and any statement that the work must be original to you. If the contract says no generative tools, the workflow ends here. Humanizing the output does not change what it is.
If the contract is silent, raise it now rather than at delivery. A two-line clause agreed up front settles what tools may do and what you must disclose. Our AI disclosure clause builder drafts one, and the longer reasoning sits in what a design contract should say about AI.
Step 1: Draft from the client's material
A humanizer can change how a sentence sounds. It cannot add knowledge the draft never had. So the draft has to start from the brief, not from a one-line prompt. Give the model the client's notes, approved product names, the audience and two or three paragraphs the client wrote themselves. Ask for plain structure, not polish; polish is the next step.
Keep the prompt and the raw output in the project folder. If a question about the copy comes up months later, that file answers it faster than memory does. Mark which sections came from the model so you know where to look hardest when you edit.
Step 2: Humanize with Walter Writes
Paste the draft into Walter Writes or upload the file; it takes output from ChatGPT, Claude or Gemini. Then make two choices.
- Strength. Start with Simple for short pieces such as product blurbs and email subject lines. Use Standard for most body copy. Save Enhanced for long passages that still read stiff after a Standard pass. The lighter the setting, the more of the original wording and the client's terms survive.
- Tone. Brand-safe suits most client copy because it is neutral enough to carry the client's voice on top. Casual fits social posts and informal newsletters. Academic fits white papers and reports.
Mind the size of each request. Starter accepts 750 words per request, Pro 1,500 and Elite 2,000, per the pricing page, so split long documents at headings rather than mid-paragraph. Paid plans start at $12 a month, or $8 a month billed yearly, for 30,000 words a month. If you already run tasks through an AI assistant, every plan includes MCP server access, and there is a REST API for heavier use.
Run the built-in detector on the draft before and after. Treat the result as a hint about which paragraphs still sound generated, not as a grade to hit.
Step 3: Edit by hand for voice and facts
This is where the fee is earned. A humanizer varies rhythm and word choice; it does not know the client. Read the result out loud once, with the client's own writing open beside it, and fix three things.
- Terms. Put back every product name, feature label and house phrase the rewrite changed. A short list of the client's preferred terms makes this a two-minute check.
- Facts. Check every number, date, price and claim against the brief. Rewrites can soften a precise statement into a vague one, or the reverse.
- Openings and endings. Machine drafts tend to open with a generic setup and close with a summary nobody needs. Cut both and start with the point.
| Step | What you do | Tool | What to keep on file |
|---|---|---|---|
| Contract | Confirm AI use is allowed and what must be disclosed | Contract, clause builder | Signed clause or written OK |
| 1. Draft | Prompt from the brief and the client's own copy | Any chat model | Prompt and raw output |
| 2. Humanize | Lightest strength that works, brand-safe tone | Walter Writes | Before and after text |
| 3. Edit | Restore terms, check facts, cut filler | You | Edited file with tracked changes |
| 4. Review | Read against the brief; detector only as a hint | You, optional detector | Notes on what changed |
| 5. Disclose | Send a short note on what was AI-assisted | Email or handover doc | The note itself |
Step 4: Review without chasing a score
Some clients paste delivered copy into a detector. It helps to know how weak that evidence is before the conversation happens. OpenAI withdrew its own AI text classifier in July 2023 because of its low accuracy; at launch it caught 26% of AI-written text and wrongly flagged human writing 9% of the time. A Stanford team led by Liang found that popular detectors consistently misclassified writing by non-native English speakers as AI-generated, which matters for the many freelancers who write for clients in their second language.
So review against the brief, not against a percentage. Does the copy say what the client needs, in words the client uses, with facts that check out? If yes, the job is done. Never promise a detector result in a proposal; promise the process instead.
Step 5: Disclose and hand over
Disclosure follows the contract. Where the contract requires it, and in our view where it does not, send a short note with the files. Three sentences are enough: which sections started as an AI draft, which tool rewrote them, and that you edited and fact-checked the result. For example: "Sections 2 and 3 began as an AI draft from your brief, were rewritten with Walter Writes, and were then edited and checked by me against your product sheet."
Keep the before and after versions with the project. That record is the same kind of evidence we describe in how to prove a design is yours, and it protects you if the copy is ever questioned.
Choosing the humanizer for step 2
We use Walter Writes in this workflow because its controls map onto the steps: strength for how much to change, tone for the register, a detector for the before-and-after check, and plans priced by the word. Two other tools are worth knowing. Grammarly can build a custom voice from a writing sample, which helps when matching a client's voice matters most, and QuillBot's Freeze words feature locks terms you do not want touched. Our comparison of AI humanizer prices per 1,000 words for freelancers covers the wider field, including limits and plans.
Frequently asked questions
How do I humanize AI text without changing the meaning?
Use the lightest rewrite setting that fixes the rhythm, then compare the result line by line with the original. In Walter Writes that means starting on Simple or Standard rather than Enhanced, and restoring any term or figure the rewrite altered.
Can I humanize AI text for free?
For short pieces, yes. Walter Writes lets you humanize 300 words free with no signup, according to its humanizer page. Regular client work will outgrow that quickly, and paid plans start at $8 a month billed yearly.
Is it OK to use an AI humanizer on client work?
Yes, when the contract allows generative tools and you disclose how you used them. It is not OK when the contract forbids AI or the client is paying specifically for original writing from you.
Do I have to tell a client I humanized AI text?
If the contract or the client's policy mentions AI, yes. If it does not, we still recommend a short note with the delivery, because it costs nothing and prevents a dispute later.
Which humanizer setting works best for client copy?
Brand-safe tone at Standard strength is a sensible default for most client copy in Walter Writes. Drop to Simple for short, precise text and move to Enhanced only for long passages that still read flat.
Will humanized text pass an AI detector?
No one can promise that. Detectors disagree with each other and misfire on human writing, as OpenAI's withdrawn classifier and the Stanford study on non-native writers show. Agree on process and disclosure with the client instead of a score.
The short version for your next job
Check the contract, draft from the client's material, humanize with Walter Writes at the lightest setting that works, edit by hand, and disclose. Walter Writes is our recommendation for that middle step because it gives you control over strength and tone, a detector for a quick before-and-after look and a free 300-word allowance to test it. The tool handles the rhythm. The judgement, the facts and the signature on the invoice stay yours.