AI training opt-out vs AI shopping visibility: a PrestaShop guide

Three separate decisions

Training concerns whether a provider may use web content to change a generative model’s learned parameters. Crawling and search concern whether a bot may fetch or surface a page. Agent commerce concerns whether an authorized assistant can read a public catalog through a declared interface such as MCP, UCP or REST. A single robots.txt setting does not settle all three.

For a PrestaShop merchant, the objective might be to express a preference against model training while keeping product pages findable and the public catalog queryable. That is a policy decision for the merchant, not a default that a plugin should silently impose.

What was announced on 30 October 2026

An individual IETF Internet-Draft, version 00, proposes a policy token named __usage_ai-training. It reuses ordinary robots.txt groups. The draft is not an approved IETF standard; its syntax and interpretation can change. It expresses a preference about use for training, not a security barrier and not a crawl prohibition.

User-agent: __usage_ai-training
Disallow: /

This illustrative configuration expresses a site-wide preference against AI training to providers that implement the draft. It does not guarantee compliance by all bots, undo past training, or block a user-authorized agent from reading a public product page. Do not paste it into a live shop without deciding the policy and testing existing crawler rules.

Provider support is not simultaneous

OpenAI says it intends to implement the shared signal in the coming months; its existing crawler controls remain in force. Google targets 2027 and says an explicit Google-Extended instruction can take precedence over the shared preference. Microsoft Bing says it will honor the new training and page-level grounding preferences from 1 February 2027, with migration actions for older NoArchive/NoCache controls by 1 March 2027.

Grounding—using live pages to support an AI answer—is a different choice from training. Microsoft warns that restricting grounding reduces visibility in AI experiences. Google also documents interactions between Google-Extended and eligibility for Gemini grounding. These rules differ by provider: there is no single “visible to every assistant” switch.

A safe checklist for PrestaShop

1. Write down the merchant’s objective: training preference, search visibility, or public agent catalog access.
2. Review existing robots.txt: named crawler rules, wildcard rules, cache/CDN behavior and any explicit Google-Extended setting.
3. Check discoverability separately: product URLs, public product data, robots/search controls and declared MCP/UCP endpoints.
4. Test as an anonymous shopper: product variants, public price, taxes, availability and restricted/B2B visibility must remain correct.
5. Record evidence and dates: save the effective robots.txt, provider documentation version and test results; revisit when the draft or provider implementations change.

Never treat a voluntary robots.txt preference as a substitute for access control. Private prices, customers and orders must remain protected by PrestaShop’s server-side authorization rules.

Where Fondouk fits—and where it does not

Fondouk Free exposes the public, read-only catalog of a compatible PrestaShop store through documented agent-facing interfaces. It does not decide the merchant’s AI training policy, grant access to private data, or guarantee traffic or sales from an assistant. The new training signal does not replace MCP or UCP.

Start with the guide on what AI assistants read from your store, then consult the developer guide to UCP on PrestaShop. To understand the protocol boundaries, read MCP, UCP and ACP: different roles. The free Fondouk module and source code can help you test public catalog discovery without changing your training preferences.