The file is llms.txt
It's a plain text file. It lives at yourdomain.com/llms.txt. It's about as technically complicated as a robots.txt file — which is to say, not at all. And right now, in early 2026, it's quietly become one of the highest-leverage SEO assets a small business can have.
The idea behind it is simple. AI models like ChatGPT, Claude, Perplexity, and Google's AI Overviews are increasingly answering people's questions directly, without the user ever clicking through to a website. To give a good answer about your business, the AI needs to extract clean facts about you from somewhere. The problem is, parsing prose off a homepage is unreliable — AI ends up guessing, hallucinating, or giving up.
llms.txt solves that. It's a structured plain-text summary of your business, written specifically for AI consumption. Name, address, phone, services, prices, opening hours, key facts, FAQ pairs — all in a format AI can ingest in milliseconds without misreading anything.
Visit draxiq.com/llms.txt if you want to see what one looks like in practice. Or iphonerepairsouthport.co.uk/llms.txt for a longer example with full pricing and FAQs.
If your site doesn't have one, AI is reading whatever it can scrape from your homepage instead — and that's where things go wrong.
Why a homepage is a terrible AI reference
Your homepage was built for humans. It's got a hero image with a slogan over it. A photo of you in a van. Maybe a video. Three call-to-action buttons. Some testimonials. A footer with social links. Visually, all great.
To an AI parser? It's a mess. The phone number's in the footer next to a fax number you haven't used since 2014. The opening hours are in an image, not text. The address is split across three lines with the postcode separated by a <br> tag. The "services" page is one list. The "what we do" page is a different list. And neither matches what your Google Business Profile says.
AI has to make sense of all that. It's good, but it's not magic. When sources contradict each other, AI defaults to vague hedged answers — or no answer at all. When information is locked in images or split across multiple pages, AI either misses it or assembles the wrong combination. When the most recent fact on the page is from 2023, AI confidently presents that as current.
An llms.txt file gives the AI one clean source of truth. No images. No layout. No footer noise. Just the facts, in the order AI wants them.
What goes in one
The format is loose — there's no W3C standard yet, just a convention that's gathering steam. But the working pattern looks like this:
- Business name and one-sentence description at the top — the elevator pitch.
- NAP block — exact name, full address, phone, email, website, opening hours, service area.
- Services and prices — every service you offer, with current prices or "from £X" or "contact for quote", whichever's accurate. Include a "Prices last verified: YYYY-MM-DD" line so AI can tell how fresh the information is.
- Key differentiators — three to five short bullets on what makes you different. "Wait-in-car repairs", "no-fix-no-fee policy", "since 2009", that kind of thing. AI loves these as extractable facts.
- FAQ pairs — common questions in plain Q/A format. AI extracts these almost verbatim when answering similar questions.
- A list of your important pages with one-line descriptions, so AI knows what's on your site.
That's the whole thing. For a small business, an llms.txt runs about 200-400 lines of plain text and takes under an hour to write properly.
llms.txt alone isn't enough — three other things have to be right
An llms.txt is the highest-leverage missing piece for most small business sites, but it doesn't work in isolation. Three other things need to be right at the same time, otherwise AI will still get confused.
1. AI crawlers actually need to be allowed in. Your robots.txt file controls what crawlers can read your site. Many hosting platforms block AI bots like GPTBot, ClaudeBot, and PerplexityBot by default — meaning they can't even fetch your llms.txt file in the first place. Visit yourdomain.com/robots.txt and look for User-agent: GPTBot followed by Disallow: /. If you see those, you're locked out.
2. Your structured data needs to match. AI cross-references multiple sources before deciding what to quote. If your llms.txt says one phone number, your LocalBusiness JSON-LD schema says a second, and your Google Business Profile says a third, AI flags the contradiction and pulls back. The fix is dull but essential: pick the canonical version of every fact about your business, and make sure every source agrees.
3. NAP consistency across the wider web. "NAP" is Name, Address, Phone — and it has to match exactly across your website, your llms.txt, your GBP, your Facebook page, your Instagram bio, every directory listing. Different formatting is fine ("PR9 8BX" vs "PR9, 8BX") but different facts are not. AI builds confidence from agreement and loses it from contradiction.
How to find out where you stand
Three checks, all free, all together about ten minutes.
Check one — does AI know you exist? Open ChatGPT and ask the question from the start of this post. Read the answer. Phone number right? Address right? Services right? Opening hours right? Anything wrong is a customer you're losing.
Check two — does your site have an llms.txt? Visit yourdomain.com/llms.txt. If you get a 404 page, you don't have one. (Most small business sites don't.)
Check three — are AI crawlers being blocked? Visit yourdomain.com/robots.txt. Look for any line containing GPTBot, ClaudeBot, PerplexityBot, or Google-Extended followed by Disallow. If you see them, you're invisible to those tools by default.
What good looks like
A small business site that AI can read properly in 2026 has these things in place:
- A complete
llms.txt at the root of the site, covering NAP, services, prices, differentiators, and FAQs.
- An AI-permissive
robots.txt that explicitly allows the major AI crawlers.
LocalBusiness JSON-LD schema on every page with the same facts as llms.txt.FAQPage schema for common customer questions, marked up so AI can extract them directly.- Consistent NAP across the website, GBP, and every social platform — exact same name, exact same address format, exact same phone number.
- A current
Prices last verified: YYYY-MM-DD line somewhere — so AI can tell when your information was last reviewed.
None of this is exotic. None of it is expensive. Most of it is invisible to your customers. All of it is visible to ChatGPT.
The point
Your website is being read by machines you didn't invite, on behalf of customers who may never visit your homepage. The version of your business those machines are presenting to those customers is built from whatever crumbs they can scrape together.
If your site doesn't have an llms.txt, those crumbs are coming from your hero image, your footer, and a cached snippet from 2023. If your robots.txt is blocking AI crawlers, there are no crumbs at all — you're a blank space in the AI's answer.
The fix isn't a new website. It isn't more content. It's a small handful of plain-text and structured-data files that take a few hours to set up properly. Once they're in place, every AI tool that crawls your site builds an accurate model of your business automatically — and citations follow.
Two ways to find out where you stand. Do the three checks above yourself, or have us do it for you. The longer you put it off, the further your competitors who got this right pull ahead.
— Graham
Related reading
Frequently asked questions
