A publisher deciding how Google may use its pages has more than one switch to consider. Google-Extended addresses specified Gemini uses. Search Console offers a separate control for generative Search. SynthID concerns identifying generated content. Treating them as one general-purpose opt-out would obscure what each can actually do.
Kent Walker’s 26 August statement sets out Google’s preferred approach to AI and intellectual property, including arguments about existing copyright principles and machine-readable choices. It is advocacy by a company with a commercial interest in the outcome. The statement does not itself change copyright law or grant a publisher permission to reuse somebody else’s work.
Google-Extended is a product-use instruction
Google’s crawler documentation describes Google-Extended as a robots.txt product token, not a crawler with its own HTTP user-agent string. Its stated scope covers content used to improve future Gemini models and to ground specified Gemini Apps and Vertex AI responses. Grounding means using source material to inform a response; it is different from training a future model.
The documentation says Google-Extended does not control ordinary Google Search inclusion or ranking. A website operator therefore needs to distinguish the AI uses it wishes to restrict from its conventional Search visibility. Blocking a differently scoped crawler or assuming the token covers every Google service would not express the same instruction.
Generative Search has a separate setting
Google’s website-owner announcement describes a Search Console control for generative features including AI Overviews, AI Mode and the relevant Discover experiences. Its dated update says worldwide rollout completed on 31 August 2026. Google says a site that opts out receives no traffic or impressions from its generative AI Search features. It says it does not use this setting as a ranking signal outside those generative features.
For an independent publication or a brand publishing substantial editorial material, that creates a specific tradeoff: restricting these uses also gives up their discovery channel. Before choosing the setting, a site owner can consider how that exposure serves the publication and its contributors. Record the property, the decision and who will review it when the publishing policy changes.
A freelancer whose work sits on a client’s website may not control that property. Their discussion belongs with the site operator and the agreed use of the work, rather than an assumption that every creator can personally change the hosting site’s settings.
A watermark answers a different question
Google DeepMind describes SynthID as embedding detectable signals into supported AI-generated images, video, audio and text. Its purpose is to help identify generated or altered material. That is a provenance signal, not evidence of the permissions behind a commission.
A positive watermark result cannot, by itself, establish a model release, an image licence or consent to depict a person. Nor should a missing signal be treated as proof that no AI was involved. For a creative team, recording where an asset came from and documenting its permitted use remain different tasks from checking whether a particular watermark is present.