# AI text detection

Source: https://iachecker.com/docs/text-detection
Status: Available now
Last verified: 2026-08-08

Signal 3.2 evaluates several writing patterns directly in the browser. English and French have calibrated readings; other detected languages are presented with more conservative confidence.

## Local browser analysis

The standard pasted-text detector runs in the browser. Pasting text does not by itself send the raw passage to IA Checker. The analysis calculates deterministic features and returns the report in the page.

- Cadence and sentence-length variation
- Punctuation distribution
- Repeated phrasing and discourse patterns
- Structural regularity and grounding cues

## What affects confidence

*Text evidence depth*

| Input condition | Typical evidence depth |
| --- | --- |
| Under 60 words or fewer than 3 sentences | Low |
| Under 200 words or fewer than 8 sentences | Moderate |
| Longer calibrated English or French sample | Good |
| Language outside current calibration | Conservative / unknown |

These bands describe the amount of measurable structure, not whether the final score is high or low.

## Sentence-level review

The annotated result helps locate the passages that contribute most strongly. Use those highlights to compare sections, identify abrupt changes and decide where a human review should focus.

> **Very short text**
>
> A title, slogan or single sentence does not contain enough rhythm and structure for a strong authorship reading. Add surrounding context whenever possible.

## Documentation index

- [IA Checker documentation](https://iachecker.com/docs)
- [Assistant index](https://iachecker.com/llms.txt)
- [Privacy Policy](https://iachecker.com/privacy)