A client says your work is AI and will not pay

This is a payment dispute, not a verdict on your writing

When a client runs your draft through a detector, gets a scary number, and freezes payment, it feels like a judgment on your craft. It is not. It is a payment dispute wearing the costume of a quality problem, and the detector is not the authority the client thinks it is.

AI detectors carry false-positive rates commonly reported in the five to fifteen percent range, and they reliably misfire on clean, efficient, on-brief commercial copy, which is exactly what good freelancers produce. The vendors themselves warn against treating a score as proof. So do not try to argue the number down with a rival detector. Answer with the one thing a detector can never show, which is your process.

Show your process, not your indignation

The freelancers who get paid in these situations are the ones who calmly produce a drafting trail. It is hard to fake and easy to recognize, and it reframes the conversation from a machine's guess to your documented work.

• ✓Google Docs or Word version history showing the piece built over multiple sessions, with edits and rewrites
• ✓Your research trail: the sources, briefs, and notes you worked from, with timestamps
• ✓The brief, outline, and any client feedback you already incorporated, showing iteration
• ✓Prior work in the same voice for this or other clients, establishing a consistent style
• ✓An offer to hop on a short call and revise any passage live, or talk through your choices

The reply that moves a client

Keep it short, professional, and free of outrage, because an angry reply reads as defensive and a calm one reads as certain. State that the work is original, that AI detectors are known to produce false positives on human writing (their own makers say so), and that you are happy to demonstrate your process.

Attach or link the version history and offer the live walkthrough, then restate the invoice terms. Most clients ease off at this point, because they were never certain, and a writer volunteering to be questioned about their own draft is not the behavior of someone who pasted it.

Protect the payment and your future rate

If the client still refuses, lean on structure rather than argument. If you are on a platform like Upwork or Fiverr, open their dispute or mediation process and submit your version history as evidence; on direct contracts, your signed agreement and delivery record are your leverage. Send a clear, dated request for payment that references the agreed terms.

Then protect yourself going forward. Take a deposit or milestone payments so no single piece is fully at risk. Add a clause to your contract stating that AI detection tools are not reliable evidence and that any dispute over AI use will be resolved by reviewing version history and drafting records, not automated scores. That one line turns a future ambush into a defined process you can actually work.

Build the record before you deliver

Version history helps, but it lives in your account, and a client who already distrusts you can wave it away as something you could have staged. The stronger move is a drafting record the client can verify without trusting you at all.

Firsthand, free at certnode.io/firsthand, does that. As you write, it seals checkpoints of your work, a cryptographic fingerprint of the text rather than the text itself, each one independently timestamped on a public verification page. You deliver the piece with a link, and the client sees the work grew over real time in real increments, checkable by them directly. It does not detect AI and does not claim to prove who typed. It gives you a verifiable history that is effortless to have when you did the work and impractical to fake when you did not, which is the honest heart of getting paid without a fight.