Myth or fact: can AI detectors tell if a text was written by AI?
The claim
“AI detectors can reliably tell whether a text was written by AI.”
Verdict: False
The claim that AI detectors can reliably tell whether a text was written by AI is false. The clearest evidence comes from OpenAI, the company behind ChatGPT: it built a detector, published its accuracy, and withdrew it on July 20, 2023, "due to its low rate of accuracy," according to the update on its announcement.
The claim
"An AI detector can tell me if a student, a writer or an applicant used ChatGPT."
Verdict
False. No widely used method lets someone prove, from the text alone, that a person used AI. A detector can give a probability. It cannot give proof, and it can be wrong about people who wrote every word themselves.
Where it came from
When ChatGPT launched in late 2022, teachers and employers wanted a way to check writing. Several tools appeared, and OpenAI released its own AI Text Classifier on Jan. 31, 2023.
The evidence
OpenAI tested its classifier on a set of English texts and published the results:
How OpenAI's AI text detector performed before it was withdrawn in 2023
In other words, the tool missed about three of every four AI-written texts, and it labeled about one in 11 human-written texts as AI-written. OpenAI also warned that the classifier was "very unreliable on short texts (below 1,000 characters)" and recommended using it "only for English text."
Six months after launch, OpenAI withdrew it. If the company that builds the model could not detect its own text reliably, a single score from any detector is weak evidence on its own.
Why is AI text so hard to detect?
A language model writes by choosing likely next words, so its text tends to be smooth and typical. But many people also write smooth, typical text, especially in formal assignments, and a person can edit AI text until it reads differently. A detector is guessing from style, and style overlaps between people and machines.
What should you do instead?
- Treat a detector score as a reason to ask questions, never as proof.
- Ask the writer about their process: drafts, notes and sources are better evidence than a score.
- In schools and workplaces, set clear rules about when AI use is allowed, so detection matters less.
FAQ
Are AI detectors accurate?
Not reliably. OpenAI's own detector caught 26% of AI-written text and wrongly flagged 9% of human writing before OpenAI withdrew it on July 20, 2023.
Did OpenAI shut down its AI detector?
Yes. OpenAI said its AI Text Classifier was no longer available as of July 20, 2023, due to its low rate of accuracy.
Can a detector prove a student used ChatGPT?
No. A detector gives a probability, not proof, and it can flag text that a person wrote alone.
Sources
Cite this
Myth or fact: can AI detectors tell if a text was written by AI?, New Mind Museum, November 6, 2026. https://newmindmuseum.com/culture/2026/11/myth-ai-detectors-tell-ai-written-text/
New Mind Museum. (2026, November 6). Myth or fact: can AI detectors tell if a text was written by AI?. https://newmindmuseum.com/culture/2026/11/myth-ai-detectors-tell-ai-written-text/
“Myth or fact: can AI detectors tell if a text was written by AI?.” New Mind Museum, 6 Nov. 2026, https://newmindmuseum.com/culture/2026/11/myth-ai-detectors-tell-ai-written-text/.
@misc{new-mind-museum-2026-myth-or-fact-can-ai-detectors-tell-if-a-,
author = {{New Mind Museum}},
title = {Myth or fact: can AI detectors tell if a text was written by AI?},
year = {2026},
url = {https://newmindmuseum.com/culture/2026/11/myth-ai-detectors-tell-ai-written-text/}
}Free to quote and cite with a link to this page.