Everyone Is Adding An llms.txt File. Google Just Told You It Does Nothing.

Right now there is a guru in your feed telling you that if you do not have an llms.txt file, ChatGPT will never mention your business. He is selling a $500 "AI visibility" package. The centrepiece is a text file you could write on a napkin.

Let me save you the money and the panic. llms.txt is fine. It takes five minutes. And it is nowhere near the thing that decides whether AI names you or your rival. The data is in, and it is brutal for the people hyping this.

Here is the honest version, with the receipts, and then the stuff you should actually spend your afternoon on.

What llms.txt actually is, in plain English

llms.txt is a plain text file you drop at the root of your site, like robots.txt or sitemap.xml. It lives at yoursite.com/llms.txt. Inside, you write a tidy, human readable map of your site: a short description of who you are, then a list of your most important pages with a line explaining each one. Some people also ship an llms-full.txt with the actual content stitched together in Markdown.

The idea, proposed by Jeremy Howard of Answer.AI back in September 2024, was neat: large language models have a limited context window, HTML is noisy, so give the model a clean summary and it will understand your site faster. On paper, lovely. In practice, the machines mostly did not ask for it.

llms.txt is not robots.txt and it is not your sitemap

People muddle these three, so let me line them up. They do different jobs, and only two of them are actually respected by the machines that matter.

See the pattern? robots.txt and sitemap.xml earned their place by being useful to the crawler. llms.txt asks the crawler to trust your self written summary instead of just reading the page. That is a big ask, and the big engines have declined it.

The uncomfortable data nobody selling it will show you

This is where the dream meets the server logs. SE Ranking ran the numbers on 300,000 domains. Two findings matter, and they both land hard.

First, adoption is tiny. About 10.13% of sites have an llms.txt file. Nearly nine in ten do not bother. And among the top 1,000 sites by traffic, adoption was basically zero. The biggest, best resourced sites on the internet looked at this and shrugged.

Second, and this is the one that ends the argument: there was no measurable citation lift. Once you control for the things that actually matter (site authority, how much schema you use, how fresh your content is), having an llms.txt file did nothing to how often AI models cited a domain. Both plain statistics and a machine learning model found no effect. None.

What Google actually said (out loud, on the record)

You do not have to trust me on this. In July 2025, Google's Gary Illyes said flatly that Google does not support llms.txt and has no plans to. Then John Mueller put the boot in with the comparison that should tell you everything: he likened llms.txt to the old keywords meta tag.

If you are new, the keywords meta tag was a box where you typed the words you wanted to rank for. Search engines ignored it for over a decade, because it is self declared. You are the one filling it in, so of course you claim you are the best plumber in three counties. A file where you describe yourself is not evidence. It is a brag. Machines learned to ignore brags a long time ago.

Google went further in its own AI features guidance, updated in mid 2026, stating you do not need to create new machine readable files, AI text files, or Markdown to show up in Google Search or its AI features, because Search itself does not use them. That is not a rumour off a forum. That is the platform telling you where to not waste your time.

So is llms.txt completely useless? No. Here is the honest bit.

I am not going to pretend it is worthless just to make a punchy point. That is what the hype merchants do in reverse. Here is the fair take.

llms.txt was never really an SEO tool. The smarter framing is that it is a business to agent play. As AI agents start actually doing things (booking, buying, comparing on your behalf), a clean machine readable surface that says "here is what this company does and here are the pages that matter" could genuinely help an agent route to the right place. Some crawlers from OpenAI and Microsoft do fetch these files. It is a small bet on where the web is heading, not a switch that floods you with citations today.

So treat it like a cheap insurance policy. Five minutes now, in case the agent web grows up and starts leaning on it. Just do not confuse an insurance policy with a growth strategy.

How to add one in five minutes, then walk away

If you want the box ticked, here is the whole job. You do not need a plugin or a $500 package.

1. Open a plain text editor.
2. Write one H1 with your business name. Under it, one short line describing what you do and who for.
3. Add a section per group of pages (## Services, ## Guides, ## About). Under each, list the key URLs as Markdown links with a short note after each one.
4. Keep it honest and short. This is a menu, not a sales letter.
5. Save it as llms.txt and upload it to the root of your site so it loads at yoursite.com/llms.txt.

That is it. If you want to be thorough, add an llms-full.txt with your core content in Markdown. Then close the file and never think about it again. We keep a copy paste template for this inside the Dojo so members do it in one sitting and move on to the work that pays.

Five ways people waste time on llms.txt

If you are going to add one, do not turn a five minute job into a five hour one. These are the traps.

• Stuffing it with keywords. It is a menu, not a ranking lever. Keyword spam here does nothing except make you look desperate to the one human who reads it.
• Paying someone to "install" it. It is a text file uploaded to your root. If a package charges hundreds for this, you are paying for the panic, not the work.
• Listing every URL you own. Point to your best, most important pages. A bloated list buries the pages you actually want an agent to route to.
• Letting it go stale. A menu that links to pages you deleted a year ago is worse than no menu. If you keep one, keep it current.
• Thinking the job is now done. This is the big one. Ticking the llms.txt box and calling your AI search strategy finished is how you stay invisible while a sharper rival gets named.

Where that saved afternoon should actually go

Here is the part the file sellers skip, because it is harder than uploading a text doc. If you want AI to recommend your business, you have to earn the mention. The models do not trust what you say about yourself. They trust what the rest of the internet says about you. That is the whole game in 2026.

Four moves beat an llms.txt file every single time:

1. Win the answer, not the ranking

In AI search there is no page two. You are named in the answer or you are invisible. Put a clean, 40 to 60 word answer at the top of every important page, directly under the question as a heading. The model lifts that first tidy block. Bury it and someone clearer gets quoted instead. We break this exact structure down in our guide on how to get AI to recommend your business.

2. Get talked about on sites the models already trust

Third party mentions are the currency. Reddit, in particular, is one of the most cited domains in AI answers. Get genuinely helpful in the right threads, get named in "best of" listicles for your niche, show up on YouTube and industry directories. This is the bit agencies charge $2,000 a month for, and it is mostly graft plus knowing the order to do it in.

3. Label everything with schema

Schema is boring and it works. FAQ, Organization, LocalBusiness, HowTo. It is the label on the tin that tells a machine exactly what it is reading. Most of your rivals cannot be bothered, which is precisely why it is an edge. Start with our 2026 schema markup guide and ship the easy ones this week.

4. Be one clear entity, said the same way everywhere

One name, one job, one description, repeated identically across your site, your socials, your directory listings and your Google Business Profile. Consistency is how a machine learns you exist as a real thing. If you say "digital marketing" here and "growth consultancy" there, you have taught the model that you are two blurry half things instead of one sharp one.

That is the real AEO stack. None of it is a magic file. All of it compounds. If you want the full system with the templates, the schema snippets and the tracker, that is what the AEO playbook and the Dojo are built around.

Who should bother with llms.txt, and who should not

The 2026 view: where this actually goes

Do not write llms.txt off entirely, and do not build your quarter around it. The realistic future is that the standard survives as a niche B2A convenience for the agent economy, useful for docs and large sites, ignored by the big search engines that keep saying they will not use it. If the agent web gets serious, your five minute file was a smart little hedge. If it does not, you lost five minutes. Either way you were never one text file away from AI stardom.

The businesses winning AI search in 2026 are not the ones with the fanciest hidden files. They are the ones that are genuinely the best, clearest answer to a real question, and have got the rest of the web to say so. That is unglamorous. It is also the only thing that has ever actually worked. If you would rather have the exact order of operations handed to you instead of guessing, we walk through the whole thing step by step inside the Dojo, and our team runs it for you through Done For You if you would rather buy back the time.

Stop chasing magic files. Start earning the mention. The full AEO system, templates, schema snippets and citation tracker live in one place. Join The Dojo. $67/mo. Cancel anytime.