Short answer
AI detectors are classifiers trained on examples of human and machine text. They estimate how predictable the wording is, how evenly sentences run and how familiar the phrasing looks, then report a probability or percentage. They make mistakes in both directions, change over time and disagree with each other, so a score is a signal, not proof.
What is an AI detector?
An AI detector is software that estimates whether a piece of text was produced by a language model. Most are classifiers: models trained on large sets of text labeled human or machine, which learn to tell the two apart and then score new text.
What signals do AI detectors look at?
Vendors rarely publish their methods, but many detectors look at signals like these. Predictability: language models tend to pick a likely next word, so detectors estimate how expected each word is. Rhythm: machine drafts often keep sentences a similar length, while people vary theirs. Phrasing: openers and transitions that appear across huge numbers of drafts.
In today's fast‑paced world, (reads as AI: Stock opener) clear communication plays a crucial role in (reads as AI: Inflated phrasing) team success.
Why do AI detectors make mistakes?
The signals are not unique to machines. Formal writing, template-driven genres such as lab reports, heavily edited prose and writing by people working in a second language can all be predictable and even. A detector sees the same surface either way.
Errors run in both directions: human text can be labeled AI, and machine text can be labeled human. Vanderbilt University cited the risk of false positives when it turned off Turnitin's AI detector in 2023.
How accurate are AI detectors?
Not accurate enough to prove who wrote a text. OpenAI withdrew its own AI text classifier in July 2023, citing its low rate of accuracy. Turnitin's guidance says the percentage on its AI writing indicator should not be the sole basis for action. Liang et al. (2023) found popular detectors misclassified essays by non-native English writers far more often than essays by native speakers.
Why do two detectors disagree about the same text?
Each detector is a different model, trained on different data with different thresholds, and most are updated without notice. Two tools can read the same paragraph differently, and one tool can score it differently from one month to the next.
What should I do with a detector result?
Treat it as one signal. If the writing is yours, keep your drafts and version history, and be ready to explain your process. If you want to see which phrasing reads as formulaic, our free check explains the patterns in your text; it does not prove authorship. TemplateBreaker's paid plans rewrite AI-assisted drafts for a more natural voice, and no plan promises a detector result.
See alsoFree check for formulaic patternsPlans and pricingFlagged as AI, but you wrote it
Questions, answered
How do AI detectors decide if text is AI-generated?
Most compare the text with patterns learned from training examples: how predictable each word is, how much sentence length varies and how often stock phrasing appears. The result is a probability, not a fact about how the text was written.
Are AI detectors accurate?
Not reliably enough to prove authorship. Detector makers themselves warn against relying on a score alone, and research has found higher error rates on some groups of human writers.
Why do different detectors give different scores?
Each vendor trains its own model on its own data and sets its own thresholds, and most do not publish how they decide. Models are also updated, so the same text can score differently in different tools or months.
Can editing change a detector score?
Edits change the signals detectors measure, so scores can move, but no edit can promise a particular result. Edit for your reader, and keep your drafts as the record of how you wrote.
Sources and further reading
- OpenAI: New AI classifier for indicating AI-written text (withdrawn July 20, 2023, for low accuracy)
- Turnitin: AI writing detection capabilities FAQs
- Vanderbilt University: why it disabled Turnitin's AI detector (2023)
- Liang et al. (2023), GPT detectors are biased against non-native English writers, Patterns
