llms.txt Is a Signpost for AI Agents, Not a Ranking Button

llms.txt is just a file that briefly tells AI what your site does. It is not a mechanism that raises your Google search rankings. This Markdown file, placed at the site root, started as a proposal to guide LLMs and AI agents to key pages ^[1], but as it got tied up with the broader shift to AI search, it has spread into the misconception that "if you make one, you'll show up in AI search." This post covers what llms.txt actually does and why search rankings and Agent Readiness should be viewed separately.

Why this misconception took hold

llms.txt started as a proposal to briefly tell LLMs and AI agents what a site does, which pages matter, and which documents to read first ^[1]. The problem is that once the file got linked to the broader shift to AI search, its meaning started to get inflated. At some point, "a file that's easy for AI to read" came to be heard as "a file that helps you show up in AI search."

From a search engine optimization perspective, the question is "does this file affect search rankings?" But the question in the agent era is different: "can an agent understand this site and carry the user through to the next action on their behalf?" The two questions look similar, but they are actually different problems.

Google's position: llms.txt is not a ranking button

Google has explained that Google Search does not use llms.txt, and stated that the file neither helps nor hurts search visibility or rankings. That makes claims like these risky.

• Creating llms.txt raises your Google search rankings
• Creating llms.txt guarantees AI Overview citations
• llms.txt is a required GEO item

From a Google Search perspective, what still matters are the basics of traditional SEO: crawlable pages, clear content, structured data, a good user experience, trustworthy information, and brand and entity signals that are confirmed consistently across the web ^[3].

Why does Lighthouse look at llms.txt?

The interesting part is Chrome Lighthouse. Lighthouse has an experimental Agentic Browsing category, and it includes an llms.txt audit.

The key is not to conflate "Lighthouse checks it" with "it's used for search rankings." Agentic Browsing is not a traditional 0 to 100 SEO scoreboard. It is closer to an experimental check of how ready a site is for machine interaction. The llms.txt audit sits in that context. It checks whether the site root has a file that gives AI agents an easy-to-read summary and key links, and it is currently treated as optional. If the file is missing and returns a 404, it can be marked N/A, but if fetching fails outright, as with a server error, it gets flagged.

In other words, Lighthouse's llms.txt item is less a "ranking boost signal" and more a check for "is there a signpost that makes the site easier for agents to understand?"

Retrieval, Citation, and Agent Readiness are different

When talking about AI search, you need to separate the stages.

Being read doesn't mean being cited. Being cited doesn't mean an agent can understand your service and act on it. Of these three stages, it is safest to view llms.txt not as something that directly changes search rankings but as a signal closer to Agent Readiness.

From SEO to GEO, and on to Agent Readiness

SEO is about getting search engines to discover and evaluate your pages. GEO looks at how brands and information are selected and represented in generative answers. GEO covers which questions bring your brand in as a candidate, which passages make it into the answer, and which sources survive as citations ^[2].

Agent Readiness goes a step further. Users may soon ask an agent things like "check whether this service fits our company" or "go ahead and request a consultation." What matters then is not just ranking. The service description has to be clear, key documents have to be linked, and the purpose of each button and form has to be obvious. Accessibility structure, layout stability, and the main action paths all start to matter too.

So it would be a shame for the llms.txt debate to stop at the effect of a single file. The bigger question is: "Was our website built only for people and search engines, or is it structured so that agents can understand it and act on it too?"

SEO, GEO, and Agent Readiness compared

Here are the three concepts in one table.

The three layers stack on top of one another. A site with broken SEO will struggle to do GEO, and a site with weak GEO gives agents too little evidence to work with. So if you single out llms.txt and treat it as "the one thing to do," you end up neglecting the SEO and GEO it depends on.

The difference in a practical scenario

Take a hypothetical B2B SaaS company. It runs a product page, a pricing page, and a blog.

From an SEO perspective, you check whether the product page is indexed well for keywords like "project management tool," whether its Core Web Vitals are healthy, and whether the internal link structure is clear.

From a GEO perspective, you check which of the company's passages could serve as evidence when AI answers the question "recommend a project management tool." What matters is whether pricing, target customers, and differentiation from competing products are laid out in clear sentences.

From an Agent Readiness perspective, you check what happens when a user asks an agent, "see if this company's product fits our team and sign us up for a free trial." Can the agent understand the buttons on the pricing page, figure out what each field in the signup form asks for, and carry through to a completed signup? If button labels are vague or the form renders only through JavaScript, the agent may get stuck halfway.

All three layers deal with the same pages, but each checks something different.

What practitioners should do now

First, don't treat llms.txt like an SEO performance metric. You can create one, but you shouldn't promise higher rankings or more AI Overview citations.

Second, get basic SEO in order first. Crawling, indexing, structured data, internal links, title and body structure, page performance, and trust signals are still the foundation.

Third, from a GEO perspective, build candidate answer passages for each question. If information like product descriptions, target customers, problem scenarios, comparison criteria, case studies, and pricing terms is vague, AI will struggle to pick it up reliably too.

Fourth, from an Agent Readiness perspective, check whether the site is "actionable." Agents need to be able to understand core flows like contact, booking, signup, purchase, documentation browsing, and plan comparison.

Fifth, see llms.txt as a signpost, not a file that replaces all of this structure. A signpost without a good building means little, but if the building is well structured, a signpost can help agents find their way.

Deploying llms.txt itself is easy

Technically, creating llms.txt is simple. You upload one Markdown file to the site root and you're done. So the genuinely hard part of this debate isn't how to make the file but understanding what it doesn't solve.

Faced with the big task of "getting ready for AI search," many teams want to knock out the easiest item first. llms.txt takes very little time and can be checked off a list, so even though its effect is uncertain, it easily creates the illusion that "we've handled AI search." That illusion is the real danger. If you think making llms.txt means you're done, the hard work of fixing basic SEO and structuring content for GEO gets pushed aside.

Wrap-up: direction as a checklist

llms.txt is not a button that raises your Google rankings today, and it isn't a cure-all file for GEO. But the debate itself is an important signal. It shows that the axis for evaluating websites is widening: from sites search engines read, to sites chosen for generative answers, to sites agents can understand and act on.

• Is it clear which questions our site answers?
• Are brand and service descriptions consistent from page to page?
• Are structured data and entity information in order?
• Are there passage-level answers AI can pick up?
• Can an agent understand our CTAs and contact flows?

The next stage of search isn't simply capturing more keywords. It's building websites that people can read, AI can cite, and agents can act on. When diagnosing with TRAIL Search, we recommend looking not only at whether you're cited but also at the action paths users can follow after the answer. llms.txt isn't the whole of that change. It's a small signpost pointing the way.

How to prepare for the next few years

Agent Readiness is at an early stage with no standardized metrics yet. So rather than trying to hit a perfect score right now, the realistic approach is to refine your structure bit by bit as things evolve. You can start by building a team habit: every time you create a new page, check not only "is it easy for people to read?" but also "can an agent understand what this button is for?"

In particular, CTA button labels, form field clarity, and the number of steps in contact and booking flows matter to people and agents alike. A flow that confuses people confuses agents too. If you remember that preparing for Agent Readiness is not much different from improving user experience, you can naturally fold this change into your existing UX work instead of treating it as a special, separate project.

You can continue with The llms.txt setup guide: how it differs from robots.txt, templates, priorities, and measuring impact, Why does ChatGPT cite listicles so often? Is format enough, and how should hierarchy work, and What is GEO (generative engine optimization)? Which pages to start with, and in what order.