AI Detector Checker reads English text of at least 100 words and returns an AI-writing signal score from 1 to 99, the zone that score falls into (Likely human-written, Uncertain or Likely AI-written), a stability range, and one sentence of evidence for the length of your text from our sealed test. The score is not the share of the text written by AI, and it is not a probability. A result alone does not prove who wrote a text.
- Read the zone, then the scoreOne score for the whole passage; the AI line is 65 at every length.
- Read the text yourselfThe tool points at nothing inside the passage; you do the reading.
- Never a verdictUse provenance and human review for any decision that matters.
On this page
Result anatomy
What the card shows, word for word
The card reads, from top to bottom: Analysis complete · V4; a ring and a number coloured by zone (grey for human, amber for uncertain, green for AI); the number of 100 with AI-writing signal score under it; the verdict, with the strength chip in the AI zones; Score stability range a–b; a strip divided 35 / 65 with the marker at your score; then three fixed pieces of text.
- The fixed sentence.
Scores of 65 or more are treated as AI-written, at every text length.
- The evidence sentence for your zone and the length of your text, generated from the sealed test. For a Likely AI-written result at 150 to 299 words it reads:
In our sealed test at 150 to 299 words, about 1 in 1,000 human texts reached 65 or more. About 98 in 100 texts from current AI models did.
- The disclaimer.
The score is not the share of the text written by AI, and it is not a probability. A result alone does not prove who wrote a text.
Under those: the word count and length band, the engine and release id, the request id, and Patterns observed in this text, which are observations about the text, not the reasons for the score.
The scale
The scale and the four zones
The score is a fixed map of the engine’s internal value, built so that round numbers mean the same thing at every text length: 10 is the median human text, 35 is the top of ordinary human writing, 65 is the frozen threshold, 95 is the median AI text. A decision lock guarantees that a text above the threshold always shows 65 or more and a text below it never does.
| Score | Zone | Chip | Read it as |
|---|---|---|---|
| 1-34 | Likely human-written | none | No strong AI-writing signal was found. About 1 in 100 AI texts of 150 words or more also landed here in the sealed test, so it is not a clearance. |
| 35-64 | Uncertain | none | Below the AI line but higher than most human writing of that length. Edited, partly drafted, formal and second-language English land here more often. Treat it as “no useful answer”. |
| 65-84 | Likely AI-written | Moderate signal | Above the threshold, but within two units of it on the internal scale. Review the text and its history before you rely on it. |
| 85-99 | Likely AI-written | Strong signal | Well above the threshold, where most AI texts in development sat. Still not proof, and it names no model. |
The stability range shows how far the displayed number moves when the source families behind the map are resampled. A wide range means the exact number is soft; the zone is decided by the frozen threshold, not by the range.
Worked example
A worked example
A real development text, written by a Claude model as a cover letter of 550 words, scored 94 with a stability range of 91–97, zone Likely AI-written, chip Strong signal, length band 300–599 words. Its card carried the evidence sentence for that band.
- Start with the zone. 94 is in the strong part of the AI zone: the text is well above the threshold for its length.
- Check the range. 91–97 stays inside the AI zone, so resampling the map would not move this text out of it.
- Read the evidence sentence. It tells you how often human texts of this length reached 65 in the sealed test, and how often AI texts did. It is a rate for the length, not a statement about this text.
- Do not infer ground truth. The result describes writing patterns. To investigate authorship, bring in drafts, revision history, notes, sources and the writer’s explanation.
Meaning
What the score is and is not
It is a position on a fixed scale
Higher scores mean a stronger match to the patterns the engine learned from AI writing. 65 always means the frozen threshold, at every length.
It is not the share of AI-written text
A score of 91 does not mean that share of the text was written by a model. The engine scores the whole passage at once.
It is not a probability
No number on the card is the chance that AI wrote the text, because that chance depends on context the engine cannot see.
Use it to ask, not to claim
“Does this passage deserve a closer human read?” is the right question. “AI wrote this” is not a claim the card supports.
Evidence
The evidence sentence
Every card carries one sentence from the sealed test of 17 October 2026, chosen for your zone and length band. These are the exact strings:
| Length band | In the AI zone | In the human zone |
|---|---|---|
| 100 to 149 words | Short text. In our sealed test at 100 to 149 words, about 1 in 100 human texts reached 65 or more. About 97 in 100 texts from current AI models did. Add more text if you can. | No strong AI-writing signal. In our sealed test at this length, about 3 in 100 texts from current AI models also scored below 65. |
| 150 to 299 words | In our sealed test at 150 to 299 words, about 1 in 1,000 human texts reached 65 or more. About 98 in 100 texts from current AI models did. | No strong AI-writing signal. In our sealed test at this length, about 2 in 100 texts from current AI models also scored below 65. |
| 300 to 599 words | In our sealed test at 300 to 599 words, about 1 in 300 human texts reached 65 or more. About 99 in 100 texts from current AI models did. | No strong AI-writing signal. In our sealed test at this length, about 1 in 100 texts from current AI models also scored below 65. |
| 600 or more words | None of the 970 human texts of this length in our sealed test reached 65 or more. About 99 in 100 texts from current AI models did. | No strong AI-writing signal. In our sealed test at this length, about 1 in 100 texts from current AI models also scored below 65. |
In the Uncertain zone the sentence is always: Not enough to call either way. The score is below the AI line (65) but higher than most human writing of this length.
Uncertain
Uncertain results
About 2 in 100 human development texts and about 1 in 200 AI development texts fell in the Uncertain zone, so it is a small zone, and it is a real outcome rather than an error or a “50/50” finding. It happens most with edited or partly drafted text, formal or templated English, second-language writing and short passages.
What to do next. Read the passage. Check it for the concrete detail, sources and voice you would expect from the writer. If you can, check a longer passage from the same document: at 100 to 149 words the card itself asks you to add more text.
Scenarios
Common scenarios
A strong signal on your own writing
It happens: about 1 in 700 human texts of 150 words or more reached 65 in the sealed test, and about 1 in 100 at 100 to 149 words. Keep your drafts; see if your text was flagged.
A low score on text you know was generated
About 1 in 100 AI texts of 150 words or more scored below 65 in the sealed test; edited or paraphrased machine text was not measured and is likely to score lower still.
A formal speech or a templated document
Modern formal speeches were flagged 1.83% of the time in the sealed test, more than any other kind of human writing we measured. Read a flag on such text with that in mind.
Two passages from the same document disagree
Each passage is scored on its own; the compare mode never averages them. Different sections can legitimately sit in different zones.
Next steps
Responsible next steps
- Teachers: ask for the draft history or a short conversation about how the piece was written. One result is never enough for a decision about a student.
- Editors: compare with the writer’s earlier work and check facts and sources exactly as you otherwise would.
- Writers: keep drafts and notes; if your own text is flagged, they are your evidence.
Frequently asked questions
Why is the AI line at 65 for every text length?
Because the map is built per length band so that the frozen threshold for that band always lands on 65. The threshold itself differs by band on the internal scale; the displayed scale hides that difference so one number means one thing.
What is the difference between Moderate signal and Strong signal?
Both are in the AI zone. Moderate (65-84) means the text is above the threshold but within two units of it on the internal scale; Strong (85-99) means it is well beyond, where most AI texts in development sat. Neither is proof.
Can a human text score 65 or more?
Yes. In the sealed test it happened to 4 of 2,863 human texts of 150 words or more and to 11 of 1,136 at 100 to 149 words. That is why the card says a result alone does not prove who wrote a text.
What is the stability range?
The 5th to 95th percentile of the displayed score across 1,000 resamples of the source families behind the map, with a minimum half-width of 3 points. It describes how soft the exact number is; it does not change the zone and it is not a probability.
Why did my result say “Checked with our previous engine”?
If the engine is restarting, the card says “The checker is restarting. Please try again in a few seconds.” No other engine answers in its place, so every result you see comes from V4.
Does the score say which AI model wrote the text?
No. The engine performs no model attribution. The model guides describe how often texts known to come from each family were caught in the sealed test, which is a different question.