Somebody has probably told you to add an llms.txt file to your website. Possibly an agency, possibly a LinkedIn post, possibly a tool that will generate one for you in exchange for an email address.
I’ve been asked about it enough times to write this down properly. The short version: the evidence for doing it is currently very thin, and the people telling you otherwise are mostly repeating each other.
Here’s what’s actually known.
What llms.txt Is Meant to Be
The idea is reasonable enough. A file at the root of your site — yoursite.com/llms.txt — containing a clean, structured summary of your content, written for AI systems rather than humans. A curated map saying “here’s what matters on this site and where it lives.”
The pitch is that AI crawlers will read it, understand your site faster, and cite you more accurately.
It’s a sensible-sounding proposal. It’s also, right now, largely aspirational.
What Google Said
Not ambiguous. In July 2025 Google’s Gary Illyes confirmed publicly that Google doesn’t support llms.txt and has no plans to. John Mueller compared it to the keywords meta tag — a reference SEOs will recognise as fairly withering, given that tag has been ignored for the better part of two decades.
That’s the company running the AI Overviews appearing on roughly half of all searches, saying on the record that this file plays no part in it.
What the Data Shows
Three findings, and they point the same direction.
Finding | The number |
|---|---|
Domains with an llms.txt file (analysis of 300,000+ domains, early 2026) | ~10% |
AI bot visits that requested llms.txt (500M+ visits, 90-day window) | 408 |
Major AI companies publicly committed to reading it in production | Zero |
That middle row is the one worth sitting with. Out of more than five hundred million AI crawler visits monitored over ninety days, four hundred and eight of them asked for the file. That is a rounding error against a rounding error.
The crawlers are visiting. They’re reading your pages. They are almost entirely ignoring the file built specifically for them.
So Why Is Everyone Talking About It?
Three reasons, and none of them is evidence.
It’s easy to write about. A concrete, actionable, technical-sounding recommendation that takes an afternoon. That makes excellent content marketing regardless of whether it works.
Nobody wants to be late. The memory of businesses who ignored mobile until it was too late runs deep in this industry. “Add it just in case” feels like cheap insurance.
It’s genuinely being used somewhere else. Documentation-heavy technical companies — Anthropic, Vercel, Cloudflare, Supabase, Mintlify and similar — have adopted it, and adoption there is real. That’s led to the reasonable-sounding but incorrect inference that everyone should.
Which brings us to the version of the argument that does hold up.
The Case That Isn’t Nonsense
The strongest argument for llms.txt has nothing to do with search rankings.
It’s that coding agents and AI development tools genuinely do consume documentation, and a well-structured llms.txt makes a technical product easier for those agents to work with. Call it business-to-agent rather than SEO.
That’s why the adoption you can see is concentrated among developer tools and API companies. For them it’s a product decision — making the thing easier for AI agents to integrate — and it makes sense on those terms.
If you sell an API, a developer tool or a technical product that engineers integrate using AI assistants, there’s a real case. It just isn’t an SEO case.
Should You Add One?
Situation | My honest answer |
|---|---|
Local service business, ecommerce, professional services | No. There’s no evidence it does anything for you. |
Content site or publisher | No. Spend the afternoon on structure and specificity instead. |
Developer tool, API, technical documentation | Possibly — as a product decision, not an SEO one. |
You have spare capacity and it costs nothing | Fine. Just don’t count it as work that moved anything. |
I’m not arguing it’s harmful. A static text file nobody reads harms nothing but your afternoon.
What I’m arguing against is the framing — presenting it as an urgent AI-readiness step when the same afternoon spent on things with actual evidence behind them would be worth considerably more.
What I’d Do With That Afternoon Instead
All of these have evidence behind them, unlike llms.txt.
Put the answer in the first sentence. Take your five most valuable pages. Give each a question-shaped heading with a complete answer directly beneath it. AI systems extract from the opening lines of a relevant passage — bury the answer and they take somebody else’s.
Replace adjectives with numbers. “Fast service” is unquotable. “Same-day callout within 30 miles, average arrival 90 minutes” is a sentence an engine will reproduce and attribute to you. Every specific you publish is a citation opportunity.
Fix your structured data. Real schema — the kind Google has documented and does use. Note that your platform may constrain this: Wix caps each block at 7,000 characters, Squarespace gates per-page injection behind its Business plan, Duda only generates it for one location, Webflow generates none at all.
Get mentioned somewhere that isn’t your site. AI answers lean heavily on third-party corroboration — reviews, directories, industry sites, communities. Being excellent on your own domain and invisible elsewhere produces a good page nobody cites.
Any one of those will do more than llms.txt. All four will do considerably more.
What Would Change My Mind
Stating this plainly so this page ages honestly.
If any major AI company — OpenAI, Google, Anthropic, Meta, Perplexity — publicly commits to reading llms.txt in production, this advice changes immediately. If crawler logs start showing meaningful request volume rather than 408 in 500 million, same thing.
Neither has happened. If either does, the file takes twenty minutes to write and you’ll have lost nothing by waiting.
That’s the actual asymmetry here. The cost of waiting is twenty minutes. The cost of treating unvalidated advice as urgent is the opportunity cost of everything you did instead of the work with evidence behind it — and that’s the more expensive mistake.
Frequently Asked Questions
Does llms.txt help SEO?
There’s no evidence it does. Google confirmed publicly in July 2025 that it doesn’t support llms.txt and has no plans to, with John Mueller comparing it to the long-ignored keywords meta tag. Monitoring of over 500 million AI bot visits across a 90-day window found only 408 requests for the file. No major AI company has publicly committed to reading it in production.
Should I add an llms.txt file to my website?
For most businesses, no. There’s no demonstrated ranking or citation benefit, and the same time spent on answer structure, specificity, structured data and third-party presence has evidence behind it. The exception is developer tools, APIs and technical documentation, where coding agents genuinely consume the file — but that’s a product decision rather than an SEO one.
What is llms.txt?
A proposed standard file placed at the root of a website containing a structured summary of the site’s content written for AI systems rather than humans. The intention is that AI crawlers read it to understand the site more accurately. Adoption sits around 10% of domains, concentrated among documentation-heavy technical companies.
Is llms.txt the same as robots.txt?
No. Robots.txt is a long-established standard that crawlers genuinely respect, controlling what they may access. llms.txt is a proposed standard offering a curated content summary, and it has no confirmed support from any major AI company. One is enforced convention; the other is a suggestion largely being ignored.
What should I do instead of llms.txt for AI visibility?
Four things with actual evidence behind them: place complete answers in the first sentence under question-shaped headings, replace vague claims with specific numbers and timeframes an engine can quote, implement proper structured data within your platform’s constraints, and build presence on third-party sources such as review platforms and industry directories that AI answers draw on for corroboration.
Want the version that actually moves something?
Book a free reservation. I’ll ask the AI engines about your category live on the call, show you who they name, and tell you what those businesses have that you don’t. No files nobody reads.
Serving all 50 US states, remote. ✉ info@shazzseo.com
Sources and Caveats
Google’s position was stated publicly by Gary Illyes and John Mueller in July 2025. Adoption figures come from an SE Ranking analysis of over 300,000 domains in early 2026. The crawler request figure comes from third-party monitoring of AI bot traffic across a 90-day window. These are independent studies rather than platform-published data, and methodologies vary.
This page reflects the position as of July 2026. It’s the kind of advice that could become wrong quickly if a major provider announces support — check for newer information before treating it as settled.
Related Reading
- AI Overviews and SEO: the 2026 data — what’s genuinely happening to traffic, and the figure most articles omit.
- MCP servers for SEO — an AI development that has actually shipped and is worth your attention.
- Best Website Builder for SEO — including how each platform limits the structured data that supports AI citation.
- Can your website builder get you cited by AI? — the schema constraints that actually decide citation.
Written by Shahzaib Ul Hassan, senior AI SEO consultant and founder of ShazzSEO. Ranking sites since 2009, and reasonably resistant to things everyone is suddenly certain about.