A bare number reads as a verdict
Writing-pattern analysis is probabilistic pattern analysis, never proof of authorship, and claims of certainty are forbidden by the model instruction itself.
Writing-pattern analysis, openly uncertain
An AI-pattern score that shows its evidence and its limits
AIWriteCheck returns an AI-pattern likelihood with an explicit confidence level, labeled signal evidence, deterministic metrics, and a disclaimer on every result. Built for writers, educators, and content reviewers who need analysis, not a verdict.
- Likelihood from 0 to 100
- Confidence: low to high
- Labeled signals with evidence
- Probabilistic, never proof
- No-store API responses
A single score is not proof of authorship
Writing-pattern analysis is probabilistic. A bare number without evidence or limits invites decisions the data cannot carry.
High-stakes decisions need more
Every result carries the disclaimer that writing patterns are not authorship and that decisions require context and process evidence.
How AIWriteCheck turns text into a bounded report
Four steps, each visible in the source and each part of the result you receive.
You paste the text
The API accepts nonempty text. Short text can reduce confidence, and the model service explains text it cannot process.
Deterministic metrics are computed
Word count, sentence count, average sentence length, sentence-length deviation, and lexical diversity are measured directly.
Patterns are weighed, not judged
The analysis considers repetition, sentence regularity, abstraction, transition density, specificity, idiosyncrasy, and natural variation, under an instruction that forbids claims of certainty.
You get a report you can copy
The result pairs an AI-pattern likelihood with an explicit confidence level, labeled signals with evidence, and the disclaimer, and the full report can be copied.
The result contract, field by field
Every successful analysis returns the same bounded structure. No hidden fields, no silent certainty.
| Field | Contract |
|---|---|
| Likelihood | AI-pattern likelihood reported from 0 through 100 |
| Confidence | Low, medium, or high, reported with the score |
| Signals | Two to six signals, each with label, evidence, effect |
| Metrics | Word count, sentence count, average length, deviation, lexical diversity |
| Disclaimer | Patterns are not authorship; decisions need context and process evidence |
| Input | Nonempty text; unsupported input or model-service failures are reported, not silently truncated |
Before you rely on it
Is this a yes-or-no authorship detector?
No. The model instruction defines the output as probabilistic pattern analysis, never proof of authorship, and forbids claims of certainty.
Can a result stand alone in a high-stakes decision?
No. Every successful result includes the disclaimer that writing patterns are not authorship and that decisions require context and process evidence.
What happens to the text I submit?
The interface states that submitted text is not retained, and analysis responses are marked Cache-Control no-store.
What length of text can I analyze?
The application requires nonempty text without imposing its own character ceiling. Short text can reduce confidence; the model service reports text it cannot process in its own error.
Can I share the analysis with others?
Yes. You can copy a report containing the likelihood, confidence, summary, signal evidence, and disclaimer.
Analyze the pattern. Keep the uncertainty.
Paste your text and receive a likelihood with its confidence, its signal evidence, and its disclaimer, in one copyable report. The interface states that submitted text is not retained.