The short version
robots.txtis a long-established standard with a published specification (RFC 9309) that major crawlers honour.llms.txtis a community proposal, not a standard. No major AI vendor has committed to honouring it as a ranking or access mechanism.- To control AI crawler access today, use
robots.txtwith the specific user agents. That is the mechanism that works. - Adding
llms.txtis cheap and harmless. Treating it as an optimisation lever is not supported by anything published.
These two files get discussed as if they were alternatives at the same level of maturity. They are not. One is a decades-old protocol with an IETF specification that essentially every significant crawler implements. The other is a proposal that emerged from the community, has a reasonable rationale, and has not been adopted as a standard by the organisations whose behaviour it is meant to influence.
That distinction matters when you are deciding where to spend an afternoon, and it is worth stating plainly because a lot of content on this subject does not.
The two files compared
The risk row is the reason to be careful with the file that actually works. A misplaced Disallow: / in robots.txt can remove your site from search results, and it happens regularly, usually when a staging configuration reaches production. Nothing comparable can go wrong with a file nobody reads.
Controlling AI crawlers properly
If your goal is to allow or block AI crawlers, robots.txt is the mechanism. AI vendors publish the user-agent strings their crawlers use, and those are what you address. The important decision is directional: search retrieval and model training often have separate agents. Decide which use you want to restrict and check that provider's current controls.
User-agent: *
Allow: /
Disallow: /api/
Disallow: /dashboard
User-agent: GPTBot
Allow: /
Disallow: /api/
Disallow: /dashboard
User-agent: ClaudeBot
Allow: /
Disallow: /api/
Disallow: /dashboard
User-agent: PerplexityBot
Allow: /
Disallow: /api/
Disallow: /dashboard
Sitemap: https://example.com/sitemap.xmlNote that user-agent strings change and new crawlers appear. A blanket User-agent: * rule covers agents you have not enumerated, which is usually what you want (enumerate specific agents only when you want to treat one differently from the default).
Should you block AI crawlers?
There is no single answer for all AI agents. Blocking a training crawler can be a different decision from blocking search retrieval. Access controls affect compliant requests, not references from other sites, historical training data or every assistant answer. Choose based on your publishing rights and distribution goals.
The exception is publishers whose entire business is monetised attention on the page. If a summary replaces the visit and the visit was the product, the calculation genuinely differs. That is a commercial judgement about your model, not an SEO decision, and it deserves to be made deliberately rather than inherited from a template.
If you do decide to publish an llms.txt, treat it as documentation rather than optimisation. A short, accurate description of what your site is and where the important sections live costs an hour and is harmless. What it will not do is influence whether you are cited, and building a workflow around maintaining it is effort better spent on the entity clarity and structured data that plausibly do.
Check what you are already blocking
A meaningful number of sites are blocking AI crawlers without knowing it, through a rule added years ago for an unrelated reason or inherited from a boilerplate configuration. Open your own /robots.txt and read it. It takes thirty seconds and it is the highest-value check on this page.