llms.txt is presented in some discussions as a way to guide models to a site’s content. Merchants should distinguish a community proposal from an official directive supported by a specific search system. Do not let an optional file replace site fundamentals.
Evaluate the convention’s scope
Before adding a technical convention, ask who documents it, which system consumes it and what verifiable behavior it triggers. Publishing a URL does not by itself prove crawlers read it, prioritize it or use it to answer.
Check official documentation for the systems that matter and avoid turning possible adoption into a universal prerequisite for AI citations.
Start with controllable foundations
Ensure priority pages are public, working, mobile-friendly and linked from site navigation. Use informative titles, visible main content and consistent canonical URLs. Maintain your sitemap and robots rules according to editorial choices.
For products, verify actual details, variants, prices and availability. Any structured data should match the visible page. These elements have documented uses and directly benefit shoppers even if no AI system uses the page.
If you publish llms.txt, measure without overpromising
Treat the file as a low-priority experiment: keep it maintainable, limit it to useful public pages and state exactly what you hope to improve. Do not remove important site content and replace it with a link list.
Compare outcomes using a defined method, but do not infer causation if several changes happened at once. Distinguish crawler access, observed mentions and actual visits.
Sources and references
Frequently asked questions
Is llms.txt required to be cited by ChatGPT or Google?
Do not treat it as a universal requirement. Check the official guidance for the system concerned; prioritize public access, useful content and documented robots controls.
Can I list all products there to make them surface?
A list does not require any system to crawl or surface those pages. Work on internal links, product-page quality and accurate structured data where applicable.