llms.txt — the smaller business’s secret weapon for AI visibility

llms.txt is a new file type that tells AI crawlers which content on your website is worth citing — and it costs 0 kr. Get a comprehensive guide on how your SME is cited by ChatGPT, Perplexity and Gemini.

llms.txt is a cheap file to create, and almost no companies have one. But we owe you an honest introduction: there is currently no evidence that the file in itself will get you more AI citations. We recommend it nonetheless — for reasons we explain below — but not as a shortcut to visibility.

We scanned 465 Danish company websites visible in search results ourselves on 21 July 2026. 89 per cent had no llms.txt file — see the full survey, including methodology and caveats. Globally, adoption stands at around 5 per cent of the one million most-visited sites (HTTP Archive analysis, June 2026). By way of comparison, AI bots accounted for 4.2 per cent of all HTML requests globally in 2025 (Cloudflare Radar Year in Review) — a real and growing figure, but not the same as traffic to Danish SME sites, for which no published source provides figures.

What is llms.txt — and why should smaller businesses take an interest in it?

llms.txt is a simple Markdown file that you place in the root directory of your domain, providing AI models with a curated overview of your most important content. The format was proposed by Jeremy Howard (Answer.AI) on 3 September 2024; the specification can be found at llmstxt.org. Whereas robots.txt tells crawlers what they must not fetch, llms.txt suggests what is worth reading.

Does it work? The honest answer

No — not in the sense that the industry often claims. Ahrefs analysed 137,210 domains in May 2026 and found that 97 per cent of all llms.txt files were never fetched even once (Ahrefs, May 2026). SE Ranking found, across ~300,000 domains, no measurable correlation between having the file and being cited in AI responses. And no AI provider has ever committed to reading it: neither OpenAI’s nor Anthropic’s official crawler documentation mentions llms.txt at all. Google’s John Mueller stated this explicitly in June 2025: no AI system uses llms.txt today.

So why do we recommend it? Because it takes less than an hour, can’t do any harm, and is cheap insurance in case the format gains traction. But it’s low-hanging fruit — not a shortcut. What actually makes a difference in our own metrics is readable text without JavaScript, structured data and accessibility for AI crawlers in robots.txt. If you only have an hour, spend it on that.

What’s the difference between robots.txt and llms.txt?

Many SME owners are familiar with robots.txt — the file that tells search engines what they can and cannot index. But llms.txt is something completely different. Whereas robots.txt tells you “what to avoid”, llms.txt tells you “what to use”.

Property	robots.txt	llms.txt
Purpose	Block unwanted crawling	Recommend the best content
Target audience	Googlebot, Bingbot	GPTBot, ClaudeBot, PerplexityBot, CCBot
Format	Technical directive syntax	Plain text — Markdown links
Impact on AI visibility	Critical — if you block the crawler, you cannot be cited	No documented effect (see the section above)
Time to implement	10–60 minutes	10–20 minutes

Note the order in the table: robots.txt is the more important of the two. An incorrect line there can make you invisible to ChatGPT — this is a measured, real-world effect. llms.txt is the optional extra. The industry has got into the habit of selling them the other way round.

Why is the AI overlooking your business right now?

The problem: AI models have a limited context window. When GPT-5 or Claude 4 needs to find an ‘auditor’, it doesn’t scan the entire internet. They use a RAG (Retrieval-Augmented Generation) pipeline, which retrieves the most relevant documents from their index. Without an llms.txt file, the model doesn’t know which pages on your domain are your strongest.

Imagine having a library with 100 books, but with no catalogue — the librarian would simply grab the first book off the shelf. The AI does the same with your website. It takes the homepage (which is often general marketing content) and misses your page on ‘accountant’s fees’, which would otherwise be the perfect answer to the customer’s question.

Note, however, that the image above is an explanation of why the file was invented — not proof that it works. As mentioned, the major studies point in the opposite direction. What actually determines whether the model can ‘read’ you is whether there is any text at all in your HTML without JavaScript.

How many people actually use AI to find businesses?

There are no Danish figures on how many people use AI to find local services — we’ve looked, and we’re not going to make up a figure. The closest relevant Danish figure: In April 2026, Dansk Erhverv found that 45 per cent of Danes had used an AI tool in connection with an online purchase within the past six months, and that one in four uses AI for recommendations. This figures relate to shopping, not local services — but it’s genuine, Danish and verifiable.

Relark.dk: From a GEO Score of 49.9 to measurable AI visibility

The result: Relark.dk — Martin’s own AI product for middle managers — achieved a GEO Score of 49.9 (F) in the first test. The four strongest dimensions were llms.txt (100%), Mobile (100%), Security (90%) and Basic Page Signals (70%). The biggest gaps: AI Crawler Accessibility (0), Structured Data (0) and E-E-A-T Content Quality (38).

• Before Geoa: GEO Score 49.9 — F — AI crawlers were unable to crawl the site
• Day 7: robots.txt updated with ‘Allow’ for GPTBot, ClaudeBot and PerplexityBot
• Day 14: Structured Data corrected and SSR implemented
• Day 28: Measurable improvement in AI crawling and indexing

Read the full case study: Relark.dk — before and after the GEO case.

How to create your own llms.txt — step by step

It takes 10–20 minutes, and you don’t need to know how to code:

1. Identify your 5–10 most important pages — Home, About Us, Pricing, Contact, Services pages, Case Studies
2. Create the file — Structure with # About and ## Links sections
3. Upload to the root folder — Place it so that it’s accessible at your-domain.dk/llms.txt
4. Test and verify — Visit the file in your browser, check it using the URL Inspection Tool

Why llms.txt is better than paid adverts

AI searches work fundamentally differently from Google Ads. When a user asks ChatGPT “Who is the best hairdresser?”, there is no ad space. There are only organic results. A llms.txt file costs 0 kr. Google Ads for “hairdresser” cost 8–15 kr. per click. At 100 clicks per month, you’ll pay 8,000–15,000 kr. per month — whilst llms.txt gives you free visibility in AI answers.

llms.txt is one signal among several. If you want the full order of play — what you do first, second and third to become visible in AI answers — it is set out in the guide to how your business gets into ChatGPT.

Frequently asked questions about llms.txt

Can llms.txt replace SEO?

No — llms.txt is a complement. SEO optimises for Google, whilst llms.txt optimises for AI models. The best strategy in 2026 is to do both.

Could llms.txt harm my current SEO?

Not at all. Llms.txt does not affect Google’s indexing. It is a separate signal that is only read by AI crawlers.

Do I need to update llms.txt regularly?

Yes — the recommendation is to update the file every time you add significant new content. Maintenance takes 2–3 minutes per update.

In short: 89 per cent of the 465 Danish sites we scanned have no llms.txt. The file takes less than an hour to create and can’t do any harm — so go ahead and create it. But don’t expect citations from it alone: 97 per cent of all llms.txt files are never fetched, and no AI provider has promised to read them. If you really want to boost your AI visibility, start by making your robots.txt file accessible to AI crawlers and ensure that your HTML contains readable text without JavaScript. That’s where the impact is measured.