AI Visibility · 26 June 2026

By Eddie Eastwood, Founder

llms.txt Explained: What It Is, and Whether a Small Business Actually Needs One

You have probably seen the term in a newsletter or a LinkedIn post, usually wrapped in a warning: add an llms.txt file to your website or AI search will leave you behind. One vendor guide goes as far as calling it "the existential risk of ignoring llms.txt".

If you run an electrical firm, a small law practice or a removals company, that lands somewhere between confusing and alarming. You have a website. Now there is apparently a file you are missing, and missing it might make you "functionally invisible" to ChatGPT.

The short answer, if you only read one line: llms.txt is harmless, quick to make, and almost certainly not the thing standing between you and an AI recommendation. What actually gets you cited sits elsewhere, and we will point you to it.

What llms.txt actually is

llms.txt is a single plain-text file that sits at the root of your website, at an address like yourbusiness.co.uk/llms.txt. It is written in Markdown, the same simple formatting used for plenty of online documents, and it gives an AI system a short, tidy map of your site: who you are, what you offer, and links to your most useful pages with a line describing each one.

The idea was proposed by Jeremy Howard, co-founder of Answer.AI and fast.ai, in September 2024 (via answer.ai, 3 September 2024). The reasoning behind it is genuinely sensible. As the proposal puts it, language models "face a critical limitation: context windows are too small to handle most websites in their entirety", and "converting complex HTML pages with navigation, ads, and JavaScript into LLM-friendly plain text is both difficult and imprecise" (llmstxt.org)^[1]. In other words, a normal web page is cluttered with menus, cookie banners and scripts that an AI has to wade through to find the actual content. llms.txt hands it a clean summary instead.

It helps to know what the original proposal was really aimed at. The first examples were software documentation: the FastHTML coding library used it so AI coding assistants could find programming references cleanly (llmstxt.org)^[1]. That origin matters, because a developer-tools file and a plumber's five-page website are very different problems, and a lot of the breathless advice quietly forgets the difference.

How it differs from files you may already know

If you have ever dealt with a website, you may have heard of two older files. A quick comparison clears up where llms.txt fits.

The plain framing: robots.txt is a lock on the door, a sitemap is a list of every room, and llms.txt is a one-page note saying "here are the three rooms worth your time". The first two are near-universal and respected by every major engine. The third is new, optional, and, as we are about to see, not yet doing much.

What the evidence actually says

There is real research on whether llms.txt changes anything, and it is worth more than any confident assertion in a sales email.

The large independent study: no measurable effect

The most useful study comes from SE Ranking, an SEO platform with no llms.txt product to sell. In November 2025 they analysed nearly 300,000 domains. Two findings stand out.

First, adoption is thin. "Out of the nearly 300,000 domains we analyzed, only 10.13% had an LLMs.txt file in place," they reported, which means "almost 9 out of 10 sites haven't adopted it yet" (SE Ranking, November 2025)^[2].

Our own field data points the same way. QBiz Leads AI's AI Visibility Gap Report 2026, a technical readiness audit of 173 UK local service business websites, detected an llms.txt file on just 53 of them, 30.6% of the sample^[5]. Nearly seven in ten of those local service sites had no such file at all. Among the small businesses this post is written for, llms.txt is still the exception rather than the norm.

Second, and more important, they tested whether having the file made a site more likely to be cited by AI. They built a machine-learning model and checked. Their conclusion was blunt: "when we removed the LLMs.txt factor, the model's predictions actually improved. It became more accurate and reliable on test data." Having the file "didn't make a domain more likely to be cited by AI models. In fact, the model performed better without it" (SE Ranking, November 2025)^[2].

Read that twice, because it is the heart of the matter. An independent analysis of 300,000 sites found no positive link between having llms.txt and being cited by AI. The file was statistical noise, not a signal.

That said, the evidence is not all one way. This large independent study found no effect, but a smaller study points the other way: CiterLabs reported that a well-structured llms.txt lifted citation rates by 15 to 40% for developer-focused B2B sites. We weigh that study, and its commercial caveats, in full below.

What the major AI platforms actually do

The platforms back this up. Google published its own guidance on succeeding in AI search in May 2025. It lists what matters: "unique, non-commodity content", a good page experience, making sure crawlers "can find them, crawl them, index them", and structured data that matches your visible content (Google Search Central, May 2025)^[3]. The whole document is the same advice Google has given for years. It does not mention llms.txt at all.

SE Ranking's research notes the broader picture across platforms: there is "no clear evidence that major AI platforms are actively using LLMs.txt in their data pipelines", and while "GPTBot sometimes fetches LLMs.txt files", "this doesn't happen often" (SE Ranking, November 2025)^[2]. One real exception is worth naming: Anthropic recommends llms.txt for sites that want to be cited cleanly by its Claude model. So it is not useless everywhere. It is just nowhere near the universal must-have it is sold as.

The vendor study that says it helps (read the small print)

In fairness, there is research pointing the other way, and a balanced guide includes it. CiterLabs, a firm that runs "GEO" (generative engine optimisation) services, tracked 1,000 well-known B2B SaaS sites. It found that sites adopting a well-structured llms.txt see "15-40% higher citation rates within 60 days" (CiterLabs, April 2026)^[4].

That sounds impressive, so apply two filters. First, CiterLabs sells the service that this finding promotes, so treat it as vendor research with a commercial interest, not neutral fact. Second, and more useful, look at the wording: the lift came from a "well-structured" file, and the sites studied were "B2B SaaS" companies, mostly "developer-focused brands" (CiterLabs, April 2026)^[4]. That is the exact audience llms.txt was invented for: software companies with sprawling documentation. It is the structure and the quality of the content doing the work, not the mere existence of the file. A tidy file pointing at genuinely useful pages helps an AI; an empty file pointing at thin pages does nothing.

So the two studies do not really contradict each other. The plain synthesis is this: for a typical small local business, an llms.txt file on its own has no demonstrated effect on whether AI recommends you. Where it helps at all, it is because it points cleanly at content that was already worth citing.

Why the hype exists (and why you can ignore the fear)

If the evidence is this measured, why does so much of the advice sound like a fire alarm?

Because some of the loudest voices are selling something. The "existential risk" framing tends to come from AI-visibility vendors whose guides warn that without llms.txt you are functionally invisible to AI answer engines, that your content will be misrepresented, and that your brand's credibility is at stake. A company that sells AI-visibility software has a commercial interest in making you feel behind. Some of the underlying points are reasonable (clean, well-structured content does help AI read you). The "existential risk" wrapper is marketing.

There is a simple tell for this kind of advice. If a piece tells you a 30-minute file is the difference between thriving and disappearing, but never mentions the independent study showing no correlation, it is selling, not informing. The reality is calmer: llms.txt is a small, sensible, low-risk addition that the major engines mostly are not using yet, on a standard that may or may not catch on.

Caring about being found in AI answers is right. Believing one file fixes it is not.

So should your business add one? A plain verdict

Here is a straight decision you can act on today, based on the kind of business you run.

If your website is small (a handful of pages): do not bother yet, or add a basic one and forget about it. A removals firm or a barber with five pages does not have a "context window" problem for an AI to solve. The file would point at the same pages the AI can already read in seconds. Your time is far better spent on what genuinely makes the difference (covered below). There is no harm in having one; there is just very little point.

If your website is larger (many service pages, a resource library, location pages, guides): a hand-written llms.txt is a reasonable, low-cost tidy-up. Think of an estate agency with dozens of area pages, or an accountancy practice with guides on VAT, payroll, limited companies and self-assessment. A short file flagging your strongest pages takes under an hour, carries no technical risk, and means that the engines which do read it (Claude today, possibly more later) get a clean map. Treat it as good housekeeping, not a growth lever.

Either way, write it by hand, do not auto-generate it. Several plugins will spit out an llms.txt for you, but most produce a bare list of URLs with no descriptions, which is the version the research found does nothing. If you make one, write a real one-line description for each page in plain language. A file that says "AI visibility audits for UK service businesses" is worth ten that just dump a link.

One firm rule for every business: llms.txt only helps if it points at pages worth reading. Point it at thin or vague pages and you have simply indexed the problem. Fix the content first.

What actually gets a local business cited by AI

This is the part the file-focused advice skips, and it is the part that matters. If you want AI tools to name your business, the evidence points consistently to a handful of fundamentals. None of them is a file.

• Clear, specific content that answers real questions. Google's own guidance keeps returning to "unique, non-commodity content" that genuinely helps a reader (Google, May 2025)^[3]. A page that plainly answers "how much does an EICR cost for a three-bed in Leeds?" is citable. A page of brochure language is not.
• A site an AI can actually read. Some website builders load their text with JavaScript that several AI crawlers do not run, leaving your pages effectively blank to them. This sits underneath everything else, and it is worth checking. We cover the self-test in our guide to whether AI can read your website.
• Consistent business facts across the web. Your name, address and phone number matching everywhere (your site, your Google Business Profile, directories) makes AI more confident about who you are.
• Reviews and genuine mentions off your own site. AI reads the wider web, not just your homepage. Reviews, trade directories and real community mentions all feed the picture.
• Structured data (schema) that matches your visible content. This is a supporting technical signal, not the heart of AEO, but Google explicitly values structured data that mirrors what is on the page (Google, May 2025)^[3]. It does more for most small sites than llms.txt does.

That list is the real work. For the full picture of how engines choose who to name, see how AI engines pick sources, and for the complete local playbook, start with our pillar guide, Answer Engine Optimisation for local businesses.

If you want a deeper, step-by-step walkthrough of writing an llms.txt file itself (the exact Markdown structure, what to include, how to keep it current), we keep that on our resource page on llms.txt rather than repeating it here.

Common misconceptions about llms.txt

• "Adding llms.txt will make ChatGPT recommend my business." No file can force any platform to recommend you. At most, a good one makes your best content easier to read. Whether you get cited depends on the quality and trust signals above.
• "Google uses llms.txt." Google's published guidance on AI search relies on its long-standing signals and does not reference llms.txt (Google, May 2025)^[3].
• "Everyone has one now, I'm late." Only about one site in ten has one, and adoption is spread evenly rather than concentrated among the winners (SE Ranking, November 2025)^[2]. You are not behind.
• "The auto-generated one from my plugin is fine." Most auto-generated files are URL dumps with no descriptions, the exact version the research found ineffective. Hand-write it or skip it.
• "It is risky to add one." It is not. The file is small and exposes nothing you would not already publish. The only real cost is the time, which is why a thin site should spend that time elsewhere.

Frequently asked questions

What is llms.txt in one sentence?

It is a short Markdown file at your website's root that gives AI tools a clean summary of your most important pages, proposed in September 2024 and still optional.

Do I need llms.txt for my small business?

Probably not as a priority. Independent research across nearly 300,000 sites found no link between having the file and being cited by AI (SE Ranking, November 2025)^[2]. If your site is large and you want to tidy it for the engines that do read it, a hand-written file is a reasonable hour's work. If your site is small, your time is better spent on content, reviews and crawlability.

Will llms.txt hurt my site if I add one?

No. It carries no technical risk and exposes nothing sensitive. The only downside is spending time on it instead of on what actually drives citations.

Does Google or ChatGPT read it?

Google's AI-search guidance does not mention it. ChatGPT's crawler has been seen fetching llms.txt occasionally but not often. Anthropic's Claude is the clearest case of a platform that recommends it. Treat broad reliance as unproven for now (SE Ranking, November 2025)^[2].

Should I let a plugin generate it automatically?

Better to write it by hand. Auto-generated files are usually bare URL lists with no descriptions, which is the version that does nothing. A short, hand-written file with a real description per page is worth far more.

If llms.txt does so little, why does everyone talk about it?

Partly genuine curiosity about a sensible idea, and partly because some firms selling AI-visibility services use the fear of missing out to win business. The plain position is that it is a minor, optional tidy-up, not an emergency.

How is llms.txt different from robots.txt and sitemap.xml?

They do three different jobs. robots.txt controls permission (which pages crawlers may access), sitemap.xml controls discovery (a list of all your pages), and llms.txt suggests what is worth reading (a curated summary of your best content). The first two are near-universal web standards respected by every major engine; llms.txt is new, optional, and not yet widely used by the AI platforms.

Where exactly does the llms.txt file go, and what goes in it?

It sits at the root of your site, at an address like yourbusiness.co.uk/llms.txt, written in Markdown. A useful one names who you are and what you offer, then lists your most important pages with a real one-line description of each. The description is the part that matters: a bare list of URLs with no descriptions is the version the research found does nothing.

Does any AI platform actually use llms.txt today?

A few, unevenly. Anthropic recommends it for sites that want to be cited cleanly by its Claude model, so it is the clearest positive case. ChatGPT's crawler has been seen fetching the file occasionally but not often, and there is no clear evidence the major platforms rely on it in their pipelines. Treat broad reliance as unproven for now, and useful mainly for the engines that do read it.

My site is large enough to want one: what is the fastest way to do it right?

Hand-write it in under an hour. List your genuinely strongest pages (your main service pages, your best guides, your key location pages) and give each a plain one-line description of what it covers, in the words a customer would use. Skip the thin or duplicate pages entirely. Point it only at content that was already worth citing, because a tidy file aimed at weak pages just indexes the weakness rather than fixing it.

Where to start

If you take one thing from this: llms.txt is real, it is harmless, and it is nowhere near the most important thing you can do to show up in AI answers. Add a hand-written one if your site is big enough to benefit. Skip it for now if your site is small. Either way, put your real effort into clear content, a site AI can read, consistent business details and genuine reviews. That is what gets you named.

If you would rather know exactly where your business stands first (which pages are clear and crawlable, which need work, and whether AI tools currently mention you at all), that is what a QBiz AI Visibility check does. We look at your real setup, tell you plainly what is ready to be cited and what is holding you back, and hand you a prioritised list. No guaranteed-ranking promises, because nobody controls the platforms. Just a plain read before you spend time or money on anything, including a file you may not need.

Get your free AI Visibility check →