Google has announced that SynthID, its tool for embedding digital watermarks into AI-generated content, now supports text generated through Gemini apps and web experiences.
Google Launches SynthID to Embed Watermarks in AI-Generated Text
This article is a translation. Read the Japanese original
While these watermarks are imperceptible to humans, they can be detected using SynthID technology. For images and videos, the system is designed to maintain quality and remain resilient against modifications such as cropping and compression.
Regarding audio, watermarks will be embedded in content generated and published via Lyria and the podcast feature of NotebookLM. Google stated that these watermarks are difficult to alter through common processes such as adding noise, MP3 compression, or changing playback speed.
Text watermarking is achieved by adjusting the probability scores during word generation. This specification is designed to be undetectable to the human eye and does not affect output quality.
The presence of a watermark can be verified by uploading images, videos, or audio to Gemini and asking about them. Additionally, a verification portal called the SynthID Detector has been provided, and testing is currently underway in collaboration with journalists and media experts.
Sources: SynthID: A tool to watermark and identify content generated through AI (HN 82pt, 74 comments) (HN Search (backfill), 2026-02-27)