You Were Told to Add llms.txt. The AI Bots Are Skipping It.

The advice making the rounds, and the news that undercuts it

For about a year, a piece of advice has been making the rounds for service businesses. Add a small text file called llms.txt to your website, the pitch goes, and the AI assistants will find you and recommend you. (llms.txt is a short file some sites place at their web address so an AI gets a quick summary of the site.) Some owners built it themselves. Some paid an agency to add it.

This month the picture changed. Google updated its own documentation under the heading "Clarifying guidance on llms.txt files," and the Google guidance now states plainly that the file is not needed for Google Search and has no effect either way. That is Google saying, in writing, that the thing you may have paid for does nothing on its platform.

A separate study went further. The team at SE Ranking checked 300,000 sites and watched the automated readers that AI assistants send to a website (the bots behind ChatGPT, Claude, Perplexity, and Gemini). Over a 90-day window they logged more than 500 million bot visits. Only 408 of those visits opened the llms.txt file directly. The same study ran both a statistics check and a machine-learning model against how often each site got named in AI answers, and found no measurable effect from having the file.

So if you skipped it, you're not behind. About 1 in 10 sites added the file over the past 18 months, and the share holds steady whether the site is big or small. The owners who built it are not getting named more in AI answers because of it. The work that gets you quoted is on your own web pages, and it costs less and takes less time.

What the AI readers actually open

Those 500 million bot visits went somewhere. The bots came to sites half a billion times, skipped the special file almost every time, and read the normal web pages a visitor sees. The pages your customers land on are the same pages the AI reads. I see the same pattern in every site I audit.

When someone asks ChatGPT or Gemini for a business like yours, the assistant pulls a handful of pages it thinks are relevant and looks for one thing: a plain-text answer it can quote back to the person. It wants a price, a service, your hours, or a clear reason to pick you, in readable words. If that answer is sitting in plain text on the page, the AI can lift it and name you (named or quoted in the answer is what people mean by "cited"). If the answer is buried inside a photo, a slideshow, or a contact form, the AI reads right past it.

Most service-business homepages bury their answers under a hero photo, a slogan, and a button. The real facts live three clicks deep, or inside an image, or behind a form. A rushed customer bounces off that page in seconds, and the AI does the same thing for the same reason.

This is also why the llms.txt pitch sounded so good. It promised a shortcut: write one file, and you'd skip the work of fixing your actual pages. The study shows the shortcut leads nowhere. The bots skip the file entirely and read the pages a customer sees on every visit. The same fix helps both audiences at once. A page that answers a customer fast is the same page an AI can quote.

The five-question page check you can run today

Open your own homepage and your top services page on your phone. Read them the way a customer would, fast and a little impatient. Then ask whether the page answers, in plain words near the top, the five things a customer types into an AI assistant.

1. What you do. Name the service in the words a customer uses, such as "same-day crowns," "Saturday emergency furnace repair," or "first-time botox." A line like "comprehensive solutions for your needs" gives the AI nothing real to quote.
2. Where you do it. Your service area in plain text. The neighborhoods, the towns, or the radius you actually cover.
3. What it costs, or what it starts at. A real range beats silence. "Crowns start at $950" gives the AI something to quote. "Pricing available on request" gives it nothing.
4. When you are open. Your hours in readable text on the page, such as Monday 8am to 5pm. Avoid burying them in a PDF or an image.
5. Why someone would pick you. One or two honest reasons a customer would care about. Evening hours, payment plans, a senior tech on every call, whatever is true.

When a page answers all five in plain text near the top, it has done the real work of AI visibility. A page that answers only two, with the rest hidden, is your Monday list. The fix is usually about an hour of writing.

A quick way to test it: read each answer out loud. If you can say it in one short sentence and a person would understand it, the AI can quote it. If you have to scroll, squint at an image, or fill out a form to find the answer, so does the bot, and the bot just leaves.

The honest caveat, and what to do

The file does have a small role. Two assistants, Claude and Perplexity, do read llms.txt, and a clean one can give a modest bump on those two. The page content is what every AI reader fetches on every visit, so it earns the first hour. A tiny side file is at best a small bonus once the pages already answer the five questions. Build it later, if at all.

The reason the page work wins is plain. The bots opened the special file 408 times out of 500 million visits, and they opened the regular pages almost every single time. You can spend an afternoon on a file the bots skip, or an hour making sure your real pages answer what a customer asks. The second one is where the names get won.

So today, open your homepage and your services page on your phone. Walk the five questions. Fix the answers that are missing or hidden, and put them in plain text near the top. Tomorrow, type your most common customer question into ChatGPT and listen for your name. If it still does not say it, give it a few weeks, since the assistants re-read pages on their own schedule, then check again.