AI transparency
BlinkWrite watermark detector
BlinkWrite adds an imperceptible statistical watermark during text generation.
The detector gives reliable results for text with at least 200 tokens. Short or edited text can give an inconclusive result.
Each detection produces a signed report. The report contains a content hash, not the submitted text.
How to interpret the result
A detected watermark supports an assessment that BlinkWrite generated the text. It is not proof of authorship or identity.
Translation, paraphrasing, and substantial editing can weaken the watermark. A negative result does not prove human authorship.
The public signing keys are available from the detector key endpoint.
See how BlinkWrite's data path compares with Apple's on-device processing and Private Cloud Compute in the BlinkWrite vs Apple Intelligence writing comparison.