Google published a policy discussion on intellectual property and AI on August 26. It referenced machine-readable controls such as Google-Extended, robots.txt and protocols available through Search Console. The article represents Google's policy position, not a universal legal rule for every AI system or jurisdiction. Its useful operational signal is narrower and more practical: content can be used for model training, answer grounding, conventional search display or licensed reuse, and those activities should not be treated as one undifferentiated permission.
For exporters building visibility in search and answer engines, this distinction matters. A website needs clear product facts to remain discoverable, yet it may also contain licensed photography, third-party charts or market-specific materials that require tighter controls. A single “allow crawlers” decision cannot express all of those interests.
One crawler decision cannot represent four purposes
Website teams traditionally asked whether a page should be indexed. AI distribution adds several separate questions. May the content contribute to foundation-model training? May it be retrieved and cited when an answer is generated? Should it remain visible in ordinary search results? Do images, video or downloadable documents carry additional licensing limits?
A control designed for one of these purposes may not govern the others. Treating it as a master switch creates two opposite risks. A team may believe it has restricted training when the configuration only affects a particular product. Or it may unintentionally remove useful buyer-facing pages from search while trying to control an unrelated AI use. The name of a protocol is not enough; the team must document what the protocol actually changes, where it applies and how that outcome is verified.
Build an asset-level use register
An exporter can create a compact content-use register for its important assets. Useful fields include the URL, rights owner, source of each major visual, training position, answer-grounding position, search-indexing position, reuse terms, geography, expiry date, owner and next review date. Product specifications, company information, editorial insights and licensed media should not inherit one default without review.
This register is not paperwork for its own sake. It gives content, legal and technical teams one traceable record. When a configuration changes, the owner can record the date, expected effect, evidence and rollback route. When a platform updates its protocols, the team can identify which assets require revalidation instead of making a site-wide change based on an assumption.
What this means for Chinese exporters
GEO should not mean opening every asset to every AI use. It is a controlled decision about discoverability, buyer comprehension and rights protection. Product dimensions, applicable standards, delivery processes and contact routes generally need to be current and easy to interpret. Supplier-licensed images, employee likenesses, third-party research graphics and region-limited claims require a separate rights check.
Exporters should also avoid presenting one platform's protocol as a complete global copyright solution. Platform policies, technical controls, contracts and local law do not have identical scope. Significant rights questions still require qualified advice. The operational goal is to make routine decisions explicit and testable while escalating legal uncertainty to the right owner.
Action checklist
- Inventory product pages, insights, downloads and media added during the last sixty days.
- Add separate fields for training, answer grounding, search display and commercial reuse rather than one generic permission.
- Confirm the real product scope of every crawler or platform control with technical and legal owners.
- After a change, test public availability, conventional search access and any AI surface that should remain available.
- Record the change date, operator, expected effect, verification evidence and rollback method.
- Preserve source evidence and update dates for important product facts so that useful content is both current and rights-aware.
Turn policy statements into testable controls
Exporters do not need to wait for every platform and jurisdiction to converge before improving governance. Each control can be written as a testable hypothesis: which use it restricts, which product enforces it, whether search is affected, who owns the decision and when it must be reviewed. Server logs, Search Console signals and direct page checks can then confirm the expected result.
That approach avoids two simplistic positions: making every asset available without limits or closing valuable evidence because the environment is uncertain. Strong GEO operations keep high-value buyer facts accessible while applying deliberate, documented controls to assets with different rights or risk profiles.
Sources
- Google, August 26, 2026, policy discussion on AI, intellectual property and innovation: https://blog.google/company-news/outreach-and-initiatives/public-policy/ai-intellectual-property-future-innovation/

