# AI transparency

BlinkWrite uses AI to draft writing suggestions. You decide if BlinkWrite inserts a suggestion.

## Text watermark

BlinkWrite adds an imperceptible statistical watermark while its self-hosted model generates text.
The watermark does not add hidden characters or visible labels.

The watermark changes the probability of model tokens. A detector can measure the resulting statistical pattern.

## Detection

Use the [BlinkWrite watermark detector](https://blinkwrite.ai/ai-transparency) to inspect text.
The detector produces a digitally signed JSON report.

The report contains:

- A SHA-256 hash of the submitted text
- The report time
- The detector and watermark versions
- The model identifier
- The score, threshold, and result
- The public signing-key identifier

BlinkWrite does not retain text that you submit to the detector.

## Result limits

The detector needs at least 200 tokens for a reliable result. Short text gives an inconclusive result.

Translation, paraphrasing, and substantial editing can weaken the watermark. A negative result does not prove human authorship.

A detected watermark is evidence that the text matches BlinkWrite's watermark. It does not identify a person or a request.

## Public signing keys

Each downloaded report contains a compact JWS signature. The detector publishes the current and retired public signing keys.

The key endpoint is <https://api.blinkwrite.ai/ai-transparency/keys>.