GeoSomebody on your team probably asked you this in the last month: should we add an llms.txt file? The honest answer depends on what you're trying to show up in, and most of the advice floating around treats it as one question when it's really two.
What is an llms.txt file?
It's a plain Markdown file at yourdomain.com/llms.txt that lists a short description of your site plus a curated set of links to your most important pages, written for AI systems instead of people. The format was proposed in late 2024 at llmstxt.org, and it's meant to work like a table of contents an AI agent can read in one pass instead of crawling your whole site to figure out what matters.
Does llms.txt help with Google AI Overviews?
No, according to Google itself. Google's Search team published AI optimization guidance in May 2026 that lists llms.txt alongside content chunking and AI-specific rewriting as tactics that don't affect AI Overviews or AI Mode. The reasoning is straightforward: those AI surfaces pull from the same index that regular Google Search ranking already uses, and Google's crawlers don't treat llms.txt as an input to that index.
This isn't a new position. John Mueller has compared llms.txt to the old keywords meta tag, Google's Gary Illyes and Amir Taboul confirmed at a Search Central event that Google isn't pursuing it, and that stance has held for over a year now. If your only goal is showing up in Google's AI Overviews, llms.txt is not the lever, according to Google's own May 2026 guidance.
Do Claude, ChatGPT, or Perplexity actually read it?
Some AI systems do, but for a narrower job than search ranking. Anthropic recommends llms.txt in its own guidance on writing for agents. OpenAI maintains one for its Agents SDK and its Agentic Commerce Protocol. And Chrome's Lighthouse 13.3, shipped in early May 2026, added an Agentic Browsing audit category that checks whether a site provides the file at all.
So the honest framing isn't "nobody uses this" and it isn't "everyone requires this" either. Both of those claims are wrong. llms.txt is real infrastructure for agent-to-agent workflows and browser agents reading your site on someone's behalf. It is not a ranking signal inside a chat interface, and it is not what decides whether ChatGPT or Perplexity quotes you in a normal answer. That work is done by your page content, your structured data, and how consistently your business shows up elsewhere.
Should your business create one?
That depends on what kind of site you run. A software company with API documentation, or a product built to be used by agents, has a real reason to spend an afternoon on this: it makes the docs easier for Claude or an OpenAI agent to parse correctly. A dentist, a restaurant, a law firm, or a local shop almost certainly does not, at least not yet.
The businesses that need to be found by AI in a "best [thing] near me" query aren't losing citations because they lack an llms.txt file. They're losing them because their robots.txt quietly blocks GPTBot, their site has no LocalBusiness schema, or their pages never answer the exact question a customer is asking. Fix those first. An llms.txt file for a ten-page local business site costs almost nothing to add later, and it will not move the needle before the fundamentals are in place.
What should you fix instead?
Before an llms.txt file, check the things that demonstrably affect whether AI engines can read and cite you at all:
- Crawler access. Confirm robots.txt actually allows GPTBot, ClaudeBot, PerplexityBot, and Google-Extended. Blocking them by accident is common and easy to miss.
- Structured data. Valid JSON-LD naming your specific business type, a Dentist, a Restaurant, a LegalService, gives engines a fact they don't have to guess at.
- Answerable content. Real customer questions, answered directly in 40 to 60 words, in text that loads without JavaScript.
- Off-site consistency. The same name, address, and description across your Google Business Profile and the directories AI engines check for corroboration.
Once those are solid, adding an llms.txt file is a reasonable, low-cost extra step, particularly if you want to be well-parsed by browser agents like the ones Chrome has been previewing. Just don't let it jump the queue.
How do you know where your site actually stands?
Checking beats guessing. GeoRoar runs a free audit that scores your public pages against a fixed 100-point rubric covering crawler access, structured data, answerability, entity clarity, and off-site presence, the things that are proven to affect AI citation rather than the ones currently being debated. You get your score and the specific checks you pass and fail in about 30 seconds. If you want the fixes generated for you, that's the $99 Fix Pack.
Can you score better than this?
Here's a sample GeoRoar scorecard. Yours takes about 30 seconds.
This is a sample. Your real scorecard is free, no sign-up.
Run your free GEO audit