Method
Statuses require a first-party provider statement describing a deployed watermark in ordinary generated text. Research papers, rumors, image watermarks, and third-party detection claims do not count as confirmation.
Independent status report
AI text watermark status report, 2026
A living, source-backed record of which major AI assistants confirm watermarking ordinary generated text. Last reviewed August 11, 2026.
| AI provider | Text status | Documented technique | Coverage |
|---|---|---|---|
| Claude | Confirmed for supported models | Statistical text watermark plus signed provenance for supported files | Claude, API, Code, Cowork, Tag, AWS, Google Cloud, Microsoft Foundry |
| Google Gemini | Confirmed | SynthID statistical text watermark | Gemini app and web experience |
| ChatGPT | Not publicly confirmed for ordinary text | OpenAI confirms research; deployed provenance documentation focuses on media | ChatGPT text status not confirmed |
| Microsoft Copilot | Not publicly confirmed for ordinary text | Microsoft documents image, video, audio, and metadata marking | Text status not confirmed |
| Grok | Not publicly confirmed for ordinary text | xAI documents image and video watermarks | Text status not confirmed |
| Meta AI and Llama | Not publicly confirmed for ordinary text | Meta documents generated-image watermarking | Text status not confirmed |
| Perplexity | No public confirmation located | No documented ordinary-text watermark located | Text status not confirmed |
| DeepSeek | No public confirmation located | No documented ordinary-text watermark located | Text status not confirmed |
| Mistral and Le Chat | No public confirmation located | No documented ordinary-text watermark located | Text status not confirmed |
Correction policy
Provider documentation changes quickly. We update the table when a provider publishes new coverage or verification details. A missing confirmation does not prove that no undisclosed system exists.