What do I do if a client accuses me of using AI?

Reply calmly and quickly, state plainly that you wrote it, and attach process evidence — version history, earlier drafts, your research notes. Offer a call to discuss the piece in detail. Address the detector's reliability last rather than first, because leading with it reads as deflection even when it is accurate.

Should I refund a client who says my work is AI-generated?

Not immediately, and not as a reflex. An instant refund reads as an admission and sets a precedent for the next dispute. If the relationship is worth keeping and the work genuinely missed the brief, a revision is the better offer. If they are using a detector score to avoid paying for delivered work, that is a payment dispute rather than a quality one.

How do freelance writers prove they wrote something?

Version history from Google Docs is the strongest single artefact, because it shows the piece being built over sessions. Add your outline, research notes, and any messages where you discussed the draft. Being able to talk through why you structured it that way is often more persuasive than any document.

Can clients tell if writing is AI-generated?

Not reliably. Most are using a free detector, which estimates statistical predictability rather than authorship and produces false positives on clear, well-edited prose. Clean professional writing is exactly the profile these tools flag, which is why competent freelancers get accused more often than careless ones.

Should I put an AI clause in my contract?

Yes, and it protects both sides. State what you do and do not use, and state explicitly that detector scores are not accepted as evidence of breach because they measure predictability rather than authorship. Agreeing the standard before the work starts prevents the argument entirely.

Is it okay to use AI for client work?

It depends on what you agreed. Using it to check grammar or tighten a paragraph you wrote is ordinary editing. Having it produce the draft you deliver as your own work is a different thing, and if your client believes they are buying your writing, that belief matters. The safe route is to say what you use up front.

A Client Says Your Writing Sounds AI-Generated

What to do when a client runs your work through a detector and asks questions — how to respond, what to show, and how to set terms so it stops happening.

The message usually arrives with a screenshot. Your client has pasted your article into a free AI detector, got back a number they did not like, and now wants an explanation.

This is a different situation from a student facing an academic panel, and most of the advice out there is written for that case. You have a commercial relationship, an invoice that may be unpaid, and a reputation that travels by referral. The stakes are different and so is the right response.

First, understand what they have

Almost certainly a free online detector, used once, on the whole document.

These tools estimate how statistically predictable text is. They do not detect authorship and cannot produce a source document the way a plagiarism checker can. Clean, clear, well-structured professional writing — which is what you were hired to produce — sits close to the statistical centre and scores as machine-like for that reason.

Which produces the situation you are in: competent freelancers get flagged more often than careless ones. Polished copy with consistent register and correct grammar is exactly the profile these tools penalise. The mechanism is set out in how AI content detectors actually work.

Your client does not know any of this. They saw 87% and reasonably concluded something.

Respond quickly, and lead with evidence

Speed matters here more than in an academic case. A client left waiting fills the silence with the worst interpretation, and tells other people.

Reply within hours if you can, and keep it short.

Thanks for flagging this — I wrote the piece myself and I'm happy to show you how.

I drafted it in Google Docs between the 3rd and the 8th; the version history is attached, and you can see the structure change partway through when I moved the pricing section up. My research notes and the sources I used are also there.

Happy to jump on a call and talk through the piece if that's useful — including the parts I found hardest and what I cut.

On the detector: these tools measure how statistically predictable writing is rather than how it was produced, so clear professional copy tends to score high regardless of who wrote it. I'd be glad to send you more on that if it helps, but the drafting history is probably the more useful thing.

Note the order. Evidence first, offer second, reliability last. Leading with an attack on their tool reads as deflection even when every word is true, and it turns a factual question into a fight.

Do not do these things

Do not refund immediately. It reads as an admission, and it establishes that a detector screenshot is enough to reverse a payment. That precedent will follow you into every future dispute with that client and anyone they refer.

Do not rewrite the piece to lower the score. Altering work after a challenge is indefensible, and if the version you defend differs from the one you delivered you have created a worse problem. This includes running it through a rewriting tool — ours or anyone's.

Do not get defensive about the tools you do use. If you used a grammar checker, say so plainly. Precise, unembarrassed disclosure of ordinary editing is far stronger than a blanket denial that later gets picked apart.

Do not send them a wall of links. One clear explanation and your evidence beats a reading list. You are trying to resolve this in one exchange, not win an argument about detection science.

If you did use AI

Be exact about what and where, and do it now rather than after being pushed.

There is a real difference between using a model to check your grammar, to restructure a paragraph you wrote, or to generate a draft you then edited. The first two are ordinary editing that most clients would not think twice about. The third is the one that matters, because your client probably believed they were buying your writing.

If you are in that third category and it was not agreed, the honest move is to say so, offer to rewrite the piece properly at no charge, and change how you work. That is a bad afternoon. Being discovered later is a lost client and a story they tell.

Our guide to when and how to disclose AI use covers where the line sits in commercial work.

Fix it in the contract

The real solution is upstream. Most of these disputes happen because nobody agreed a standard, so a detector score becomes the standard by default.

Add a clause to your contract or proposal. Something like:

AI use. I use spelling and grammar tools, and may use AI assistance for research and structural feedback. All copy delivered is written and edited by me. AI detection tools estimate statistical predictability rather than authorship and produce false positives on professionally edited writing; detector scores are therefore not accepted as evidence of breach. If you have concerns about a deliverable, I will provide drafting history on request.

Three things this does. It states what you actually do, so nothing can be discovered later. It sets the evidentiary standard before there is a dispute. And it signals that you have thought about this, which most freelancers have not.

Adjust the first sentence to be true of you. A clause that misdescribes your practice is worse than none.

Keep the evidence as a habit

The artefact that ends these conversations has to exist before the conversation starts.

Draft in Google Docs. Version history is automatic, timestamped, and shows composition happening over time. Generated text does not have that shape — it arrives whole.

Never paste a finished piece in from elsewhere. A whole document appearing in one action produces a history indistinguishable from pasted generated output. Draft in place, or paste in sections as you write.

Keep the brief, your outline, and your notes in the same folder as the deliverable.

The full method is in how to keep a writing audit trail. It takes no extra effort and it converts an accusation into an attachment.

When the client is acting in bad faith

Sometimes this is not a genuine concern. A detector screenshot is a convenient way to avoid paying for delivered work, and it happens.

The tells: the accusation arrives when the invoice is due rather than when the work was delivered; they will not get on a call; they reject version history without engaging with it; or they ask for a refund rather than a revision.

If that is the pattern, stop arguing about detection. You have a payment dispute. Send a clear final response with your evidence attached, state that the work was delivered as briefed, and follow your usual process for late payment. Keep everything in writing.

And take the lesson into your next contract: deposits up front, and the AI clause above.

The part worth remembering

You are unlikely to be the last freelancer this client challenges, and you are certainly not the first. Detector scores on professional copy are high by default, because professional copy is clear and conventional — the two qualities these tools were built to penalise.

Handle it with evidence, a calm tone, and a contract clause, and it becomes a five-minute exchange rather than a lost relationship.