What to Do If a Client Says Your Writing Was AI-Generated

Your client flagged your writing as AI-generated. An AI detector scored your article, report, or web copy as "likely AI" — and you wrote every word yourself. Now the conversation is no longer about the work. It is about trust.

For freelance writers, this can affect payment, repeat contracts, and reputation. For non-native English speakers, the risk feels even sharper: polished English, careful phrasing, or a translated-and-edited structure can look "too clean" to detection tools that were never designed with multilingual writers in mind.

The goal is not to argue with a detector score. The goal is to respond professionally, show your process, and protect the relationship.

Why human writing gets flagged as AI

The perplexity and burstiness problem

AI detectors do not understand your intent, your research process, or how you revised a draft. They measure two statistical properties: perplexity (how predictable each word choice is) and burstiness (how much sentence length varies). Text that scores low on both looks like machine output to the detector — even when a human wrote it.

Who gets flagged most

This creates problems for writers whose work is:

• Carefully edited and structurally consistent
• Written in English as a second language
• Translated or adapted from notes in another language
• Polished with grammar or rewriting tools
• Following a content brief with specific keywords and structure

A Stanford study found that seven major AI detectors incorrectly flagged over 61% of human-written non-native English essays as AI-generated. The cause is structural: ESL writers naturally produce low-perplexity prose because they use safer word choices and standard grammar patterns. To a detector trained on perplexity, clean English from a non-native speaker is indistinguishable from machine output.

A false positive does not prove misconduct. But your client may not know that. They may only see a report that says "likely AI-generated." So your response needs to reduce risk for both sides.

Step 1: Do not reply defensively

A rushed or emotional reply makes the dispute worse. Avoid anything that sounds like:

"That tool is garbage. I obviously wrote it."

Instead, lead with professionalism:

"Thanks for flagging this. I wrote the piece myself, and I am happy to share the process evidence behind the draft."

This keeps the conversation on evidence. You are not dismissing the client's concern. You are redirecting it toward something verifiable.

Step 2: Ask what tool they used

"AI detected" is not enough information. Different detectors produce different results on the same text — sometimes wildly different. Ask for:

• The detector name and version
• The score or report
• Which sections were flagged
• Whether the full draft or only excerpts were tested

You can write:

"Could you send the detector report or let me know which tool was used? I want to respond to the specific concern rather than guess."

This matters because detector accuracy varies dramatically. Turnitin's own documentation acknowledges an uncertainty band of roughly plus or minus 15 percentage points depending on text length, and states the indicator "should not be the sole basis for punitive action." The University of Waterloo formally discontinued Turnitin's AI detection after finding it flagged human-written text as 100% AI in multiple instances. A single detector score is not a verdict — it is a probability estimate with a wide error margin.

Step 3: Gather your authorship proof

Your strongest response is a paper trail. Collect everything that shows the work was built incrementally by a human:

• The original brief from the client
• Your outline or content plan
• Research notes and source links
• Draft timestamps and version history
• Revision messages or comments
• Exported intermediate drafts
• Screenshots of your writing environment

If you wrote in one language first and then refined the English version — a common, legitimate workflow for multilingual writers — include the original-language draft. A client or administrator looking at a paper trail that shows "here is my outline in Portuguese, here is the English translation, here are five rounds of revision" is seeing authorship evidence that no detector can refute.

Step 4: Send a calm, structured response

Here is a template you can adapt:

Hi [Name], thanks for sharing the concern.

I wrote this piece myself. AI detectors can produce false positives — especially on polished or non-native English writing — so I am sharing the process evidence behind the draft.

Attached are the outline, research notes, draft history, and revision notes. I am happy to walk through any section you want to review.

My goal is to make sure you feel confident using the work.

Notice what this does: confirms authorship without attacking the client, explains detector limits briefly, provides concrete evidence, and keeps the relationship intact. The tone is "here is what happened" — not "how dare you."

Step 5: Prevent the next accusation

If writing is your income, authorship documentation should be part of your workflow — not something you scramble for after a dispute.

Build it into your process:

• Keep outlines and notes. Even rough bullet points in a Google Doc with timestamps are evidence.
• Save draft versions. Google Docs preserves full version history by default. So does Notion, Word, and most modern editors.
• Avoid ephemeral writing environments. If you draft everything in a chat window that disappears, you have no trail.
• Track major edits. When you restructure a section or rewrite a paragraph, note why.
• Keep client briefs and revision requests. These prove the work was iterative.
• Document your bilingual workflow. If you draft ideas in your native language and publish in English, keep both versions.

This is especially important for non-native English speakers. The bilingual workflow — think in one language, write in another, revise until the English is clean — is legitimate and common. But if a client questions the final English draft, you need to show how it was created.

The bigger picture: AI detection and ESL writers

The false-positive problem is not a niche edge case. Courts and universities are already responding. In January 2026, a New York state court ruled against a university that relied solely on an AI detector score, calling the decision made without valid basis. Multiple universities — including Waterloo, Vanderbilt, and MIT — have formally disabled AI detection tools, citing unreliability and bias against non-native English speakers.

For freelancers, the stakes are different from students but the mechanism is the same: clean, careful English gets punished by tools that equate predictability with automation. The defense is also the same: evidence of process.

If your writing is regularly reviewed by clients who use AI detectors, read the companion articles on why AI detectors misread non-native English and the legal cases reshaping this space. For a step-by-step guide to building authorship evidence, see how to prove your writing is human-written. And if your English sounds "too clean," read how to make your writing sound natural, not like AI.

How Diglot helps

Diglot is a bilingual writing workspace for people who think in one language and publish in English. Its Authorship Certificate is designed for exactly this scenario.

Every edit, paste, and AI-assist in Diglot is tracked and signed with a cryptographic key. The result is a public verification URL that anyone — your client, an editor, a content manager — can open in any browser to see a tamper-proof record of how the document was actually written. No Diglot account required to verify.

For freelancers, this turns the "did you really write this?" conversation into a link:

"Here is the verification page for this draft. You can see every edit, every revision, and the full writing timeline."

That is a different kind of conversation. One that starts with evidence, not panic.

Learn how the Authorship Certificate works, or start a free draft to try the bilingual writing workflow.