Probably not. Google's own Search team says an llms.txt file does nothing for AI Overviews or AI Mode, while Google's own Chrome Lighthouse now scores your site on whether you have one. For most B2B SaaS teams, that contradiction is not worth a backlog ticket yet.
The short version
- llms.txt is a proposed Markdown file at your site root that hands AI systems a curated index of your content. Jeremy Howard of Answer.AI proposed it in September 2024, and it is still a convention, not a standard anyone is required to read.
- Google Search's official position, published in May 2026, is that you do not need llms.txt for its generative AI features. John Mueller compared it to the old keywords meta tag.
- Chrome's Lighthouse 13.3 added an Agentic Browsing category that audits whether you have an llms.txt file, so the file now appears as a line item in a tool your engineers already run.
- No major AI system is known to read llms.txt to rank or cite you. The one real use is narrow: helping an agent that is already on your site complete a task.
- For a standard B2B SaaS marketing site, the better investment is clean HTML and legible content, not a second Markdown copy of your site maintained for bots.
Google Search and Google Chrome disagree, and that is the whole story
The confusion around llms.txt is not really about the file. It is about two Google teams pointing in opposite directions. In May 2026, Google Search Central published a guide to optimizing for generative AI features that grouped llms.txt with content chunking, AI-specific rewriting, and special schema as tactics you do not need. Days earlier, the Chrome team shipped Lighthouse 13.3 with a new Agentic Browsing category that checks your site for an llms.txt file. Same company, two answers.
It gets more pointed than guidance. Late in 2025, an llms.txt file briefly appeared on Google's own Search Central documentation and was removed within hours, while similar files stayed live on developer.chrome.com and web.dev. The most plausible read is that a shared content system deployed them, not that the Search team changed its mind. The lesson for a planning conversation is simple: do not treat the presence of an llms.txt file on a Google property, or a Lighthouse check, as Google Search endorsing it. Those are different products with different goals.
What llms.txt actually is, and what it is not
llms.txt is a plain Markdown file you place at the root of your domain, for example at /llms.txt. It is a human-curated map: a short description of your site, then a list of your most important pages with a sentence each, and optionally full-text Markdown versions of those pages so a model can read them without parsing your HTML. You can see the original proposal and format if you want the specifics.
It is not robots.txt, and it is not a sitemap. robots.txt controls what crawlers are allowed to access. An XML sitemap lists URLs so search crawlers can find them. llms.txt does neither. It is a self-reported summary, which is exactly where its problem starts. Nothing in the file is verified against your actual pages, and nothing stops every competitor from writing the same confident claims about themselves.
Why Google says it does nothing for discovery
Mueller's argument is about trust, not effort. On a mid-2026 episode of the Search Off the Record podcast, he said that an LLM system "by design" cannot trust a self-reported file as a way of differentiating between websites. If your llms.txt says you are the authoritative source on a topic, so does everyone else's. A signal that every site can assert about itself carries no ranking information, which is the same reason the keywords meta tag died twenty years ago.
The second half of the argument is adoption. Google has said no AI service it runs requests the file, and crawlers were not fetching it. A file that nothing reads cannot help you, no matter how carefully you write it. That is the state of play for search and for the big consumer assistants as of late 2026.
Why Lighthouse flags it anyway, and what that score means
Lighthouse is not grading you on rankings here. The Agentic Browsing category, still marked as under development, measures whether a browser agent that lands on your page can understand and operate it. It bundles four checks: a well-formed accessibility tree, layout stability (your existing Cumulative Layout Shift score under a new heading), WebMCP form annotations, and the presence and basic quality of an llms.txt file. The question it answers is "can an agent use this page," not "will this page get cited."
This matters for how you read the report. The llms.txt audit does not penalize you for not having the file. A missing file returns "not applicable" rather than a failure, and a plain site with no AI features can still score full marks. The first time I opened a Lighthouse 13.3 report for a SaaS client and saw llms.txt sitting in the audit list next to CLS and accessibility, the instinct in the room was to file a ticket. We did not, because a gray "not applicable" is not a problem to fix. If you let a new audit category drive your backlog, you will spend Q1 building something Google Search has already told you to skip.
The narrow case where llms.txt is worth building
Mueller left one door open, and it is the useful part. Once an agent is already on your site, a machine-readable instruction file can help it follow your structure and complete a task, such as navigating your docs or working through a checkout. That is an agent-usability feature, not a discovery feature. If your product is developer documentation, a public API, or an application that AI agents are meant to drive, scoping a small llms.txt or the emerging WebMCP annotations is a reasonable experiment with a clear owner and a clear success metric.
For the typical B2B SaaS marketing site, that case does not apply. No agent is completing a transaction on your pricing page on a user's behalf today, so a curated Markdown index buys you nothing you can measure. Keep it on the watch list, not the build list.
What to put on the 2027 backlog instead
This is the season teams lock their technical SEO roadmaps, so here is where the same hours actually pay off. Google's own AI guidance is blunt that optimizing for its AI features is, for the most part, still SEO. That means clean, semantic HTML an LLM can parse without a translation layer, fast Core Web Vitals, accurate structured data for the entities you genuinely own, and content that answers a buyer's question in the first paragraph rather than burying it. If you want the broader playbook for earning citations in AI answers, I wrote that up in GEO for B2B SaaS.
Put llms.txt in the "watch, do not build" column. Revisit it in two quarters. If a major assistant announces it actually reads the file as a citation input, that is the day it earns a ticket, and it will take an afternoon to generate. Building it now, on the strength of a Lighthouse line item, is work you cannot tie to pipeline, rankings, or conversion.
What this does not cover
This piece is about Google's AI features and Chrome's agentic browsing audit. It does not settle whether ChatGPT, Perplexity, or Claude read llms.txt, because the public evidence there is thin and keeps changing. It does not cover how to handle AI crawlers in robots.txt, which is a separate access and licensing decision. And it is not a full generative engine optimization strategy on its own. Treat it as the answer to one narrow question: should llms.txt be on your near-term backlog.
Frequently asked questions
Does ChatGPT or Perplexity use llms.txt?
As of late 2026, there is no reliable public evidence that major AI assistants read llms.txt to decide what to cite. Some crawlers have occasionally been observed requesting it, but none have confirmed it as a ranking or citation input. Treat any claim that it boosts your AI visibility as unproven.
Is llms.txt the same as robots.txt?
No. robots.txt tells crawlers what they are allowed to access. llms.txt is a proposed content summary written for LLMs. One is an access rule enforced by crawlers, the other is a self-reported index that nothing is required to read.
Will a missing llms.txt hurt my Lighthouse score?
No. Lighthouse's Agentic Browsing audit marks the check "not applicable" when no file exists and does not penalize the absence. A clean site with no AI-specific features can still score full marks in that category.
When should a B2B SaaS company actually add one?
When you run developer documentation, a public API, or an application that AI agents are meant to operate, and you want to give an on-site agent a machine-readable map to work from. For a standard marketing site focused on rankings and pipeline, skip it.
If you want a second opinion on what belongs in your 2027 technical SEO backlog, and what is just noise from a new Lighthouse category, book a 20-minute backlog read-out and I will walk your team through it line by line.