Skip to content
ChatGPTDetect
English

Can ChatGPT detect its own writing?

No. As of 2026-11-12, OpenAI's own Help Center says ChatGPT has no "knowledge" of what content could be AI-generated, and that its answers to questions such as "did you write this?" are "random and have no basis in fact." If a text matters, asking the chatbot is the least reliable way to find out who wrote it.

What OpenAI says

OpenAI's guidance for educators, updated in late October 2026, makes three separate points. ChatGPT will sometimes make up responses to questions like "did you write this [essay]?" or "could this have been written by AI?" OpenAI's research on detectors did not show them to be reliable enough for judgments about students with lasting consequences. And when OpenAI tried to train its own detector, it labeled human-written text such as Shakespeare and the Declaration of Independence as AI-generated. The guidance is at help.openai.com.

This fits the situation: a pasted text carries no record of where it came from. The same sentences could have come from a person, from ChatGPT or from another assistant, and the model has nothing to check them against.

What the research found

An October 2025 arXiv paper, Know Thyself? On the Incapability and Implications of AI Self-Recognition by Bai, Shrivastava, Holtzman and Tan, tested ten language models. The abstract reports that self-recognition was rarely better than chance and that the models tended to guess GPT or Claude as the author. It tested research settings, not the current ChatGPT app, so treat it as supporting evidence, not as a measurement of today's product.

Why a "yes" or a "no" proves nothing

Because OpenAI calls the answers random, a chatbot's "yes, I wrote that" and its "no, that looks human" carry the same weight: none. That cuts both ways. A teacher cannot use a reply to accuse a student, and a student cannot use one to prove an essay is clean. A dedicated check at least gives a measured likelihood and shows its evidence in the text.

The AI detector page inside ChatGPT

OpenAI also hosts an AI detector page among its ChatGPT writing tools, described as analyzing writing for AI-generated patterns. We found no published accuracy figures for it. OpenAI's educator guidance, above, says detectors have not been reliable enough for decisions about students, so treat any answer from it as one signal.

What does give a signal

Method What it can say Limit
Ask ChatGPT "did you write this?" Nothing dependable OpenAI calls the answers random
textGrain watermark detector Whether an OpenAI watermark is present EU ChatGPT and Codex text only; detector limited to approved researchers; student-work checks outside scope
Spotting phrases by eye A hint Humans use the same phrases; see how to tell if ChatGPT wrote it
Style-based detector Likelihood of AI writing and similarity to model families Not proof of authorship

OpenAI announced textGrain on 5 October 2026. It states that the absence of a detected watermark does not prove human authorship. Current status is on the ChatGPT watermark page.

What a careful reader does instead

Teachers will find the full workflow in AI detection for teachers. If the text may have come from Claude instead, ClaudeDetect runs the same check tuned to Claude's style.

Check a text

Paste at least 100 words into the ChatGPT detector for one free check. It reports whether the text reads as AI-written, marks the passages that carry the signal and shows writing-style similarity to ChatGPT and other families. Veread, Fileverity's text model behind it, wrongly flagged 0.5% of human-written texts in seven languages and detected 98.3% of fully AI-written texts; the facts page lists how these were measured. That is a style reading, not proof.

Sources: OpenAI Help Center educator FAQ, OpenAI, "Our approach to EU text provenance rules" and OpenAI Help Center provenance article, read on 12 November 2026; arXiv 2510.03399 (3 October 2025).

Questions and answers

Can ChatGPT tell if it wrote a text?

No, not reliably. OpenAI's Help Center says ChatGPT will sometimes make up answers to "did you write this?" and that these responses are random and have no basis in fact.

Why does ChatGPT sometimes say it wrote my text?

It produces a plausible answer to the question rather than looking up a record of its output. A 2025 research paper on ten language models found that their self-recognition was rarely better than chance.

Is there a way to check ChatGPT text that really works?

Two signals exist. OpenAI's textGrain watermark covers ChatGPT text in the EU only and its detector is restricted, while a style-based detector gives a likelihood for any text. Neither proves who wrote it.

Can I paste a text into ChatGPT's own AI detector page?

OpenAI hosts an "AI detector" page among its ChatGPT writing tools, but OpenAI's educator guidance says detectors have not been reliable enough for judgments about students. Use any result as one signal.