AI Detection False Positive: What to Do, in Order
First: this happens to people who did nothing
False positives are a documented weakness of style-based detection, not a rare glitch. Writing that is genuinely formulaic reads as generated: technical reports, lab write-ups, legal boilerplate, exam-style essays, and the work of anyone composing in a second language. Published studies have found elevated false positive rates in exactly these groups, and vendors acknowledge the limitation while still shipping the score.
So the first thing to get straight is that being flagged is not evidence that you did anything. The second is that how you respond matters more than whether the tool was right.
The order to act in
1. Do not rewrite first. Your instinct will be to change the text until the number drops. That is the one move that destroys the evidence you need later — and it makes the prose worse. Take a breath and preserve the version that was flagged, exactly as submitted.
2. Save your version history now. Google Docs, Word, Notion, Overleaf — whatever you used, export a dated copy and check that the edit history is intact. A timestamped trail from outline to final is the strongest thing you can bring to any conversation, because it is not something a model can argue with.
3. Find the specific sentences. Open the report and read which passages were flagged. This tells you what to talk about. A conversation about "seven sentences in section 3" goes very differently from a conversation about "my essay scored 80%" — and it shows you actually engaged with the report.
4. Check the settings before you argue. Was the reference list excluded? Were quotes excluded? Was the correct language supported? A surprising number of high scores come from configuration rather than content, and an instructor can check this in seconds.
5. Ask, in writing, before the deadline. Short, factual, no outrage:
I saw the AI detection score on my submission and wanted to flag it before it becomes an issue. The essay is my own; I have the document history from the first outline to the submitted version, and my reading notes. I would be glad to talk through any section you want to check.
That message does three things: it puts a timestamp on your objection, it signals you have evidence, and it offers a conversation rather than a defence.
What actually persuades people
Not the argument that detectors are unreliable. That may be true, and it is also what everyone says. What persuades is evidence of process: an outline, a reading list, a draft with comments from a friend, the version where the argument was still wrong. The messiest evidence is the most convincing, because generated text does not produce a history of getting things wrong and fixing them.
What not to do
- Do not run the text through "humanisers" and resubmit. It changes the score without changing the situation, and a second submission with a lower score and worse prose invites more questions, not fewer.
- Do not lead with the tool's error rate. It is a real issue and it is the wrong opening. Lead with your drafts.
- Do not delete anything. Old drafts, chat logs with a study group, the file on your desktop from two weeks ago — all of it helps.
- Do not post about it publicly while it is unresolved. It rarely helps and it can be read as a strategy.
If your institution treats the score as a finding
Ask two questions in writing: which threshold was applied, and what the appeals process is. Most institutions have one, and most have not thought carefully about what a style score should be worth as evidence. Asking calmly is not aggressive; it is the correct procedural move, and it puts your objection on the record.
The longer-term fix
If your writing consistently trips these tools, the cause is almost always the same: long sentences, heavy nominalisation, and transition phrases used to hold paragraphs together instead of ideas. That is worth fixing for its own sake. Varying sentence length and naming concrete things makes the prose better and, as a side effect, less machine-like. It does not make you invisible to a detector — nothing reliably does — but it makes the writing yours.
Frequently asked
Can a false positive be reversed? The score itself usually cannot be edited after submission. What can change is how the institution treats it, which is why the evidence and the written objection matter.
Should I rewrite and resubmit? Only if your instructor asks you to. Volunteering a rewrite signals that something needed fixing.
Is it worth appealing? If the score is being treated as a finding, yes — on procedural grounds (thresholds, evidence, process), not on the grounds that the tool is bad.
Will this follow me? Institutional records differ. Ask what is stored, in writing.