First: this happens to people who did nothing wrong

AI detectors flag human writing every day. Independent research keeps confirming it: a Stanford-led study found mainstream detectors flagged more than 60% of essays by non-native English speakers [Liang et al., 2023], a 2026 peer-reviewed test measured one major academic detector at 0.61 overall accuracy [2026 study], and detectors have famously flagged the U.S. Constitution as AI-generated [reported widely, 2023–2026]. Formal structure, careful grammar, a polished second draft — the very habits good writers are taught — are exactly what these tools mistake for machine output. An accusation built on a score alone is built on a known failure mode.

The playbook: what to do, in order

  1. Stay calm and write nothing angry. You are about to win on process, and everything you send may be read by a committee later.
  2. Screenshot the accusation. Save the exact message, the named tool, the score, and the date. Vague claims shrink under specifics.
  3. Pull your version history now. Google Docs (File → Version history) and Word (with AutoSave) record the messy, hours-long timeline of real writing. This is the single strongest evidence of authorship that exists — far stronger than any detector verdict, in either direction.
  4. Gather the rest of the paper trail: outlines, notes, sources you saved, earlier drafts, browser history from writing sessions, and prior graded work in your voice.
  5. Ask for the specific evidence. Request, politely and in writing: which tool, which score, and what the institution's documented policy says a score means. Many accusations dissolve at this step, because the honest answer is “a probability from a tool with a known error rate.”
  6. Cite the research. The false-positive studies in the sources list below exist to be cited. So do the universities that publicly disabled AI detection after false accusations. You are not arguing “trust me” — you are arguing the published numbers.
  7. Escalate calmly if needed. Academic-integrity processes and HR processes both have appeal steps. Bring the file of evidence, not the emotion.

A reply you can adapt

“I wrote this myself, and I'd like to resolve this with evidence. I can share my full version history and drafts showing the document being written over time. Could you tell me which detection tool produced the flag and what score it reported? Published research shows these tools have significant false-positive rates — particularly for formal and non-native writing — so I'd ask that we look at the process evidence together rather than rely on the score alone.”

Accused at work, not at school?

The same playbook applies — with one addition. Workplaces rarely have a written detector policy, so ask what standard is being applied and whether it was communicated before your work was checked. Your edit history in shared documents, commit logs, or CMS revisions serve the same role as a student's version history. And research on freelance platforms shows suspicion lands unevenly — polished, formal writers get accused more, not less [Cornell, 2025] — which is one more reason to move the conversation from vibes to process.

Check yourself first — with honest numbers

Before you reply, it can help to know what a calibrated detector actually says about your text. Paste it into the Cobalynx checker: you'll get a probability with a published error rate, not a verdict — and if the honest answer is “inconclusive,” that is what it will say. A dated result from a tool that publishes its false-positive rates is a useful exhibit; a screenshot from a tool that claims certainty is not. If you're wondering why detectors disagree with each other in the first place, we wrote up the honest numbers.

What Cobalynx can and can't do here

Cobalynx cannot prove you're human — no tool can, and any tool claiming to certify text as “human-verified” is overclaiming. What we provide: a calibrated second opinion with published error rates you can cite, an explanation of exactly how the score is produced, and this guide to the process evidence that actually demonstrates authorship. The strongest proof you have was never going to come from a detector — it is the version history you already own.

If you're on the other side of this

Teachers and managers reading this page while deciding whether to accuse someone: the fair-process version of detection — what a score can support, what it never can, and how to keep false accusations near zero — is laid out in our guide for teachers. More general questions are answered in the FAQ.

Sources

  1. Liang, Yuksekgonul, Mao, Wu, Zou, “GPT detectors are biased against non-native English writers,” Patterns (2023), arxiv.org/abs/2304.02819 — 61.3% average false-positive rate on TOEFL essays across seven detectors.
  2. Peer-reviewed 2026 evaluation of a major academic AI detector measuring 0.61 overall accuracy and a false-positive rate above the vendor's advertised figure (accessed Aug 2026).
  3. Widely reproduced detector false positives on the U.S. Constitution and other historical texts; see the overview in Wikipedia, “Artificial intelligence content detection” (accessed Aug 2026).
  4. Cornell University research on bias in who gets suspected of AI use on freelance platforms (2025; accessed Aug 2026).
  5. Weber-Wulff et al., “Testing of detection tools for AI-generated text,” International Journal for Educational Integrity (2023) — all 14 tested detectors below 80% accuracy, arxiv.org/abs/2306.15666.

Common questions

Can a detector score alone get me failed or fired?

It should not, and increasingly it does not: a detector score is a probabilistic signal with a measured error rate, not proof, and several universities have disabled AI detectors after documented false accusations. If a score is the only evidence against you, ask in writing for the specific tool, its published false-positive rate, and a review of your process evidence.

Does Grammarly or heavy editing make writing get flagged as AI?

It can. Grammar-tool polish pushes prose toward the uniform, formal patterns detectors associate with AI text, and polished non-native English is flagged at especially high rates. That is a false-positive mode, not evidence of cheating — which is why version history matters more than any score.

If I check my essay myself before submitting, am I safe?

Not guaranteed: detectors are not calibrated to each other, so a low score here does not force another tool to agree. What a Cobalynx check gives you is an honest calibrated probability with a published error rate on the evidence page — a documented, dated data point for your defense, not false comfort.

What should I say to my professor or manager first?

Stay calm and factual: state plainly that you wrote the work, offer your version history and drafts, and ask what specific evidence triggered the concern. Requesting the process — rather than protesting the verdict — moves the conversation onto ground where the documented detector error rates work in your favor.