Quick answer: AI detectors estimate how likely it is that a text was written by AI, based on patterns in the writing. They can catch a lot of unedited chatbot text, but they also flag human writing and miss edited AI text, so no detector is reliable enough to prove who wrote something. This guide brings together our research on how detectors work, how accurate the major ones are, and what to do if your writing is flagged.
Key takeaways
- A detector score is a probability, not proof.
- Every major detector, including Turnitin, GPTZero, Grammarly and QuillBot, makes mistakes.
- Formal writing, short texts and writing by non-native English speakers are flagged most often.
- Your drafts and version history are the best evidence if you are accused.
- You can check your own writing for free before you submit it.
How AI detectors work
Most AI detectors are classifiers: models trained on large sets of human-written and AI-written text so they learn the difference. When you paste in a text, the detector looks for patterns that are common in AI writing, such as predictable word choices, very even sentence lengths and a smooth, generic style, and returns a score for how AI-like the text looks.
That approach has a built-in weakness. Plenty of human writing is also predictable and even, especially formal, technical or simple writing, and AI text stops looking like AI once someone edits or rephrases it. That is why every detector produces both false positives (human writing flagged as AI) and false negatives (AI writing that passes as human).
How accurate are the major AI detectors?
| Detector | What the company says | What independent research found | Our full guide |
|---|---|---|---|
| Turnitin | Under 1% false positives for documents with more than 20% AI writing; about 4% of highlighted sentences may be human-written | Ranked first of 14 tools at 76% accuracy in a 2023 study (Weber-Wulff et al.); Vanderbilt turned it off over false positive concerns | Turnitin accuracy |
| GPTZero | 96.5% accuracy on mixed documents; false positive rate of no more than 1% | 54% accuracy, ranked 8th of 14, in the same 2023 study | GPTZero accuracy |
| Grammarly | Ranked first on the RAID benchmark with 99% accuracy | The RAID authors found detectors in general are easily fooled by edited and adversarial text | Grammarly accuracy |
| QuillBot | 99% detection rate, citing RAID | We found no peer-reviewed study of QuillBot’s detector on its own | QuillBot accuracy |
Company figures come from each company’s own published statements; independent findings come from the studies listed under Sources. Detectors are updated often, so treat any single number as a snapshot. For the big picture, see are AI detectors accurate?
Who gets falsely flagged most
- Non-native English writers. A Stanford study found seven popular detectors flagged most essays by non-native English writers as AI.
- Formal and technical writers. Predictable, impersonal wording looks AI-like to a detector.
- Short texts. A paragraph or two gives a detector little to go on, and Turnitin does not score submissions under 300 words.
- Heavily polished writing. Grammar and rewriting tools can smooth out the quirks that make writing look human.
Which AI detector does your school use?
It depends on the school. Learning platforms such as Canvas have no built-in AI detector; schools connect a separate tool, most often Turnitin. See does Canvas detect AI? for what teachers can and cannot see, and check your syllabus or ask your instructor which tool your course uses.
What to do if you are flagged
- Stay calm and ask which tool was used and what the score was.
- Show your version history. In Google Docs, go to File > Version history > See version history.
- Bring your notes, outlines, sources and earlier drafts.
- Offer to explain your work in your own words.
- Point to the research: independent studies show every detector makes mistakes.
❌ The detector says 90% AI, so I am in trouble.
✅ The detector shows AI-like patterns, so I will share my drafts and version history and explain how I wrote it.
Make your writing sound like you
If your own writing keeps getting flagged, the fix is usually more of your own voice, not a trick. Our list of AI words to avoid shows overused words with plain alternatives, and the free naturalness check in our AI humanizer highlights robotic phrasing, repetition and very long sentences so you can revise them. If your school or employer has rules about AI, follow them and be open about how you used it.
Check your writing for free
Paste your text into the free SpellMistake AI detector to see which sentences read most like AI. Treat the result like any detector score: a second opinion that helps you spot generic sentences, not a verdict.
All our AI detection guides
- Are AI detectors accurate? What the research shows about false positives, missed AI text and non-native writers.
- How accurate is Turnitin’s AI detector? Turnitin’s own false positive figures, its independent ranking and its limits.
- Is GPTZero accurate? GPTZero’s claims compared with independent tests.
- Is Grammarly’s AI detector accurate? What its benchmark result means for a real essay.
- Is QuillBot’s AI detector accurate? Its claims, free limits and what independent research does not cover.
- Does Canvas detect AI? What Canvas can and cannot see, and which detector your school may use.
- AI words to avoid Overused words that make writing look machine-made, with plain alternatives.
Related tools and guides
Frequently asked questions
How do AI detectors work?
They are models trained on human-written and AI-written text. They look for patterns common in AI writing, such as predictable wording and even sentence lengths, and give a score for how AI-like a text looks.
Which AI detector is the most accurate?
It depends on the test. In one 2023 study of 14 tools, Turnitin ranked first at 76% accuracy, but no tool reached 80%, and companies’ own figures are much higher than independent results.
Can AI detectors be wrong?
Yes. Every detector produces false positives and false negatives. Formal writing, short texts and writing by non-native English speakers are the most likely to be misread.
Can a teacher prove I used AI with a detector?
A detector score alone is not proof. It should be weighed with your drafts, version history and how well you can explain your work.
Is there a free AI detector?
Yes. The SpellMistake AI detector is free and highlights the sentences that read most like AI.
Sources
- Weber-Wulff et al.: Testing of detection tools for AI-generated text (International Journal for Educational Integrity, 2023)
- Liang et al.: GPT detectors are biased against non-native English writers (2023)
- Dugan et al.: RAID benchmark (ACL 2024)
- Vanderbilt University: Guidance on AI detection (August 16, 2023)
- Company figures: Turnitin (May 14, 2023 update), GPTZero, Grammarly and QuillBot published statements, as cited in each of our full guides
