For most websites, no. Google Search's May 15, 2026 AI optimization guide explicitly tells site owners that llms.txt is not needed for AI Overviews, AI Mode, or any other generative AI Search feature. The same guide groups llms.txt with content chunking, AI-specific rewriting, and special schema as tactics that do not help with Google AI visibility. For agent-facing developer documentation sites, API references, and properties where Anthropic Claude or OpenAI agents are a meaningful referrer, the answer is different. Anthropic explicitly recommends llms.txt in its Writing for Agents guidance, OpenAI uses it for the Agents SDK and the Agentic Commerce Protocol, and Chrome's Lighthouse 13.3 added an Agentic Browsing audit category in early May 2026 that checks whether sites provide the file.

The piece below covers what changed in May 2026, the actual empirical evidence on who uses llms.txt today, a clear decision framework for whether your site should create one, what to do if you already have an llms.txt file, and how to write one that is actually useful if you decide to keep it.

What changed in May 2026

Two contradictory updates landed inside Google within two weeks of each other. Together they produced the most confusing public guidance the SEO industry has had on llms.txt since the file was proposed in late 2024.

Google Search's May 15, 2026 AI optimization guide listed llms.txt among tactics site owners can ignore. The guide's reasoning is that llms.txt is not used by Google's Search systems, that AI Overviews and AI Mode pull from the same Google Search index that classic ranking uses, and that the file therefore has no effect on visibility inside Google's AI surfaces. The same position has been stated repeatedly by Google's Search team for over a year. John Mueller called llms.txt comparable to the keywords meta tag, noted that no AI services used it and bots did not request the file, and described building separate Markdown pages for bots as a poor use of time. Gary Illyes and Amir Taboul confirmed at Search Central Live Deep Dive Asia Pacific that Google was not pursuing llms.txt. Our full breakdown of the May 15 guide sits in our explainer on whether GEO is still SEO.

Days earlier, Chrome shipped Lighthouse 13.3 with a new Agentic Browsing category that checks for llms.txt as part of a four-audit set covering WebMCP integration, agent accessibility, layout stability, and llms.txt handling. The Lighthouse documentation describes llms.txt as a way to provide "a machine-readable summary of a website's content, specifically designed for LLMs and AI agents," and notes that without the file, "agents may spend more time crawling the site to understand its high-level structure and primary content." The audit marks sites as Not Applicable on a 404 response and flags server errors when the file cannot be retrieved.

The split makes more sense once you see the underlying logic. Google Search is talking about Search visibility. Lighthouse is talking about browser-agent readiness, specifically how well a site can be parsed by autonomous browsing agents like the Chrome-side agent capabilities Google has been previewing. The two product teams are answering different questions, and neither team is wrong inside their own scope.

The empirical reality of who uses llms.txt today

The most commonly repeated framing in the SERP is "no AI system uses llms.txt." The framing is wrong by 2026. Three pieces of evidence directly contradict it.

Google itself indexes between 30,000 and 60,000 llms.txt files globally, per research published by Wix Studio AI Search Lab using Google Advanced Search testing in October 2025. The figure fluctuates within that range across two-week sampling windows, with the vast majority of results being valid llms.txt markdown pages. Pages cannot be indexed without being crawled, which means Google is crawling these files even if its Search team says they are not used for ranking.

Anthropic explicitly recommends llms.txt in its Writing for Agents guidance. OpenAI maintains llms.txt files for the Agents SDK and the Agentic Commerce Protocol. Perplexity has been observed surfacing llms.txt content independently of standard retrieval-augmented generation. Each of these is a major AI platform actively using llms.txt as part of agent or agent-to-agent workflows, regardless of whether Google Search uses it.

The accurate framing in 2026 is that llms.txt is not a Google Search lever, but it is a legitimate agent-to-agent and agent-readiness signal that several major AI companies actively use. The binary "useless" position no longer holds up against the evidence. Neither does the universal "you must create one" position, since Google Search confirms it has no effect on AI Overview or AI Mode citation.

Should I create an llms.txt file? A decision framework

Whether to create an llms.txt file depends on what you are optimizing for. Four scenarios where llms.txt is worth the effort, three where it is not.

When to create an llms.txt file

Create one if you run developer documentation, API references, or technical documentation. Anthropic recommends llms.txt for documentation sites because agents reading the docs benefit from a curated map of what is canonical and what is navigation. The investment is small and the agent-readability gain is real.

Create one if Anthropic Claude or OpenAI agents are a meaningful referrer or use case for your site. The Claude desktop app, the OpenAI Agents SDK, and the Agentic Commerce Protocol all reference llms.txt as part of their workflows. If your traffic logs show meaningful Claude or OpenAI agent activity, or if your product is built to be consumed by agents (API products, integrations, MCP servers), the file is worth maintaining.

Create one if you are building for the agentic browser layer specifically. Lighthouse 13.3's Agentic Browsing audit is forward-looking infrastructure for browser-based agents like Chrome's preview agent capabilities, Project Mariner, and Comet. Sites that want to be cleanly parsable by these agents should support llms.txt as part of broader agent-readiness work. The investment also supports the four other audits in the Agentic Browsing category (WebMCP integration, agent accessibility, layout stability).

Create one if your site has fewer than 1,000 pages and you can maintain the file without burdening your CMS workflow. The cost of maintenance is low at small scale, and even if Google Search ignores it, the file does no harm and may produce upside on agent surfaces. The honest position for small B2B SaaS sites, indie publications, and developer-led products: yes, create one. The cost is one afternoon.

When to skip llms.txt

Skip it if your only goal is Google Search visibility, including AI Overviews and AI Mode. Google's documentation is explicit. The file does not affect citation in Google's AI surfaces, and time spent maintaining it would be better spent on the work Google does count: non-commodity content, technical SEO health, semantic HTML, and structured data for rich results.

Skip it if your site has 10,000+ pages and the maintenance cost is real. At ecommerce or enterprise content scale, keeping the llms.txt file accurate as the site changes is genuinely expensive. The agent-readiness upside does not justify the operational cost unless agent traffic is a defined business priority.

Skip it if your team is already stretched on higher-impact AI search work. The biggest mistake in 2026 is treating llms.txt as a substitute for the work that actually drives AI citation: entity-consistent content, direct-answer paragraph structure, third-party platform presence on Reddit and YouTube and Wikipedia, and citation tracking across the major AI surfaces. Our breakdown of the AEO and GEO tracking tools comparison for B2B SaaS covers the work that has measurable impact. An llms.txt file is not a substitute.

Comparison: what each lever actually does

The clearest way to think about it is a side-by-side comparison.

The pattern is consistent. The work that affects citation across all AI surfaces simultaneously (direct-answer content, third-party presence, technical health) earns the largest return per hour. llms.txt is a narrower lever specifically for agent-readiness and a subset of Claude and OpenAI agent workflows.

What to do if you already have an llms.txt file

If your site already publishes an llms.txt file, four practical options.

Keep it if your site falls into any of the four "create one" scenarios above. Maintenance involves keeping the linked pages accurate as the site changes, refreshing summaries when key pages are updated, and removing obsolete URLs. Most small and mid-sized sites can review the file quarterly.

Delete it only if you have specific evidence it is causing problems, which is rare. The most cited concern is that an outdated llms.txt could feed agents stale information, which is a real risk for fast-moving sites. If the file is not actively maintained, deletion is cleaner than leaving stale content live.

Stop maintaining it if your team is stretched and llms.txt does not pass the cost-benefit check. The file can remain as-is without active updates, and most agents that read it will still benefit from the static map of canonical pages. Resume maintenance later if your priorities shift.

Update it to align with current best practice if you decide to keep it. The current convention emerging from Anthropic's guidance and the llmstxt.org specification: place it at the root of the domain (yourdomain.com/llms.txt), use Markdown format, keep sections short with 10 to 50 curated links rather than 500, prefer .md mirrors of canonical pages where possible, and link canonical content pages rather than marketing landing pages.

How to write an llms.txt file that is actually useful

If you decide to create or keep an llms.txt file, four operational principles separate a useful file from a hollow one.

First, treat it as an editorial document, not a sitemap. The point is to curate which pages on your site are worth an agent reading first, not to mirror your full URL inventory. Twenty curated links typically outperform two hundred uncurated ones.

Second, prefer Markdown mirror pages over HTML landing pages. Most agents read llms.txt links faster and more reliably when the destination is structured Markdown. For sites with content management systems that do not natively output Markdown, the simplest fix is server-side rendering of canonical content as .md alongside the HTML.

Third, keep descriptions short and literal. Each link should have a one-line description that uses the exact terminology a buyer or agent would search for. Marketing copy in descriptions reduces agent confidence, which is the opposite of the file's purpose.

Fourth, refresh the file when your content strategy shifts. Stale llms.txt files feed agents outdated maps, which damages the trust agents have in your site over time. A quarterly review is the minimum maintenance cadence for sites that publish regularly.

Don't let an experimental file replace the work that actually moves AI citations

The most expensive mistake in 2026 is treating llms.txt as a shortcut to AI visibility. Google Search is explicit that the file does not affect AI Overviews or AI Mode. ChatGPT and Perplexity citation depends primarily on content structure, entity consistency, and third-party platform presence, not on a curated Markdown index at the root of your domain. The brands that compound AI visibility through the rest of 2026 are doing the harder work: producing non-commodity content, earning third-party citations on the platforms AI surfaces actually read, and tracking citation share across multiple AI platforms to see what is working.

The cleanest first step is a baseline audit that maps how your brand currently shows up across ChatGPT, Perplexity, Gemini, AI Overviews, and Claude. Look at how Passionfruit's GEO service approaches the audit on top of a solid SEO foundation, see the cross-platform citation tracking inside Passionfruit Labs, and talk to the team before the next budget cycle locks in.

Frequently asked questions

The questions below come up most often when teams are deciding whether to create or maintain an llms.txt file in 2026.

What is an llms.txt file?

An llms.txt file is a Markdown document placed at the root of a website (yourdomain.com/llms.txt) that provides a curated summary of the site's content for large language models and AI agents. The format was proposed in late 2024 and is described at llmstxt.org. The file typically contains a short description of the site, followed by a list of curated links to canonical pages with one-line descriptions for each.

Does llms.txt help with Google AI Overviews?

No. Google Search's May 15, 2026 AI optimization guide explicitly states that llms.txt is not needed for AI Overviews, AI Mode, or any other generative AI Search feature. Google has maintained this position consistently for over a year. The file does not affect citation or ranking inside Google's AI surfaces.

Do ChatGPT and Perplexity use llms.txt?

OpenAI publishes llms.txt files for its Agents SDK and the Agentic Commerce Protocol. Anthropic explicitly recommends llms.txt in its Writing for Agents guidance. Perplexity has been observed surfacing llms.txt content independently of standard retrieval. The file is therefore relevant for some agent-to-agent and agent-readiness workflows on these platforms, though it is not a primary ranking signal inside their consumer-facing chat interfaces.

What does Chrome's Lighthouse 13.3 check for in llms.txt?

Lighthouse 13.3 added an Agentic Browsing audit category in early May 2026 that includes an llms.txt check. The audit marks sites as Not Applicable on a 404 response and flags server errors when the file cannot be retrieved. The Lighthouse documentation describes llms.txt as an "emerging convention" and frames it as an optional but recommended file for browser-based agent readiness.

Should ecommerce sites create an llms.txt file?

Most large ecommerce sites should skip llms.txt because the maintenance cost at 10,000+ pages exceeds the agent-readiness upside. The work that actually moves AI citation for ecommerce brands is structured product data in Google Merchant Center, third-party platform presence on Reddit and YouTube, and content engineered for AI retrieval on category and comparison queries. For very small ecommerce sites under 1,000 pages, creating an llms.txt file is cheap enough to do as a small experiment.

How often should I update my llms.txt file?

Quarterly is the minimum cadence for sites that publish regularly. Stale llms.txt files feed agents outdated maps of the site, which reduces the trust agents place in the source over time. If the team cannot commit to quarterly review and updates, deletion is often cleaner than maintaining a static file that drifts out of alignment with the live site.