llms.txt: should your tech site have one?

llms.txt has become one of those topics where the hype outpaces the evidence. We’ve been asked about it in pretty much every B2B tech client meeting since late 2024. The honest answer in April 2026 is “it costs almost nothing to add, but do not expect it to do heavy lifting on its own”.

This piece walks through what llms.txt actually is, who’s adopted it, what we believe it does today and what it does not. We’ll keep the marketing claims out and stick to what we’ve seen on real client sites.

What llms.txt is

llms.txt is a proposed convention, originally floated by Jeremy Howard in 2024, that suggests a plain-text file at the root of your domain (/llms.txt) containing a curated map of your most important content for large language models. A companion /llms-full.txt includes the actual content of those pages in a clean, prompt-friendly format.

It sits alongside /robots.txt and /sitemap.xml as a way of communicating with crawlers, but it speaks specifically to LLMs that want to understand or cite your site. The format is markdown, with sections and links rather than the directives you find in robots.txt. There is no Working Group, no formal standards body and no agreed verification mechanism. It is a community proposal that has been picked up by some sites and ignored by others.

What it is not

llms.txt is not a ranking signal in any documented sense. None of the major commercial LLMs, ChatGPT, Claude, Perplexity, Copilot or Gemini, has publicly committed to honouring it as input to citation or grounding. Some crawlers respect it as a discovery aid. None of them have said “include this and your pages will be cited more often”.

It is also not a substitute for a sitemap, structured data or clean HTML. If your pages are slow, badly linked or thin, llms.txt cannot fix that. We’ve covered the deeper foundation work in our technical SEO audit checklist for tech sites and our piece on structured data for AI search.

Who’s adopting it

Adoption through 2025 was mostly developer-tooling and documentation sites. Anthropic added one. Stripe’s docs added one. Several open-source projects ship them. By early 2026 we’ve seen a slow trickle into B2B SaaS, particularly companies whose buyers research heavily through ChatGPT and Perplexity.

Among our own client base, adoption has been mixed. Some teams have shipped one as part of a broader AI SEO tidy-up. Others have looked at it and decided the time is better spent on schema, content rewrites or genuine third-party visibility work. Both positions are defensible.

The case for adding one

There is a low-cost case worth taking seriously. If you publish documentation, glossaries, comparison pages or anything else where you want LLMs to have a clean, canonical reference to your content, llms.txt gives you a place to point them. It also gives you somewhere to hand-curate which pages you’d want quoted, separate from the much larger sitemap that includes legal pages, blog archives and so on.

For tech companies whose buyers genuinely use ChatGPT or Perplexity to evaluate vendors, the file functions as a small but real signal that you take this surface seriously. We have not seen it cause any harm. We have seen it function as a useful internal forcing function: writing the file forces a marketing team to decide which thirty pages on their site they actually want represented in an AI answer, which is a healthy exercise on its own.

The case against rushing

The case against is mostly opportunity cost. If your team has limited bandwidth, llms.txt is not where we’d start. We’d start with the work that has stronger evidence behind it:

• Content rewrites that read well to a model. Covered in writing content that AI search engines actually cite.
• Schema work that supports retrieval and disambiguation. Covered in structured data for AI search and schema markup for SaaS websites.
• A third-party visibility programme. Mentions, links, podcast appearances, analyst briefings.
• Citation auditing so you actually know what’s happening. Covered in auditing your visibility in Copilot and ChatGPT.

If you’ve already done that work, adding llms.txt is a sensible tidy-up. If you haven’t, it’s a distraction.

What a good llms.txt looks like for a tech site

If you decide to ship one, keep it tight. We’d suggest the following structure for an MSP or B2B SaaS site:

# Acme Cloud Services

> Acme Cloud Services is a UK managed service provider offering Microsoft 365 management, Azure infrastructure and 24/7 helpdesk for mid-market businesses.

## Services
- [Microsoft 365 Management](https://example.com/services/microsoft-365)
- [Azure Infrastructure](https://example.com/services/azure)
- [24/7 Helpdesk](https://example.com/services/helpdesk)

## Comparisons
- [Acme vs Competitor A](https://example.com/compare/acme-vs-a)
- [In-house IT vs MSP](https://example.com/insights/inhouse-vs-msp)

## Reference
- [Pricing](https://example.com/pricing)
- [Security and certifications](https://example.com/security)
- [About Acme](https://example.com/about)

A few principles we’d apply:

• Curate, do not enumerate. Twenty to fifty links, not five hundred.
• Lead with definitional content. What you do, who you serve, what you’re known for.
• Include comparison and pricing-shaped content. These are the queries LLMs surface most for B2B buyers.
• Mirror the file in /llms-full.txt with the rendered text if you want to make it easy for a model to ingest the content directly without crawling each page.

For larger sites where a single file becomes unwieldy, we’ve written separately about structuring a more complex llms.txt.

How to monitor whether it’s doing anything

Once you ship one, treat it as an experiment, not a fix. Some of the things we’d watch:

• Cloudflare or server logs for known LLM user agents (GPTBot, PerplexityBot, ClaudeBot, Google-Extended) hitting /llms.txt. We cover the wider tracking question in tracking AI search traffic.
• Citation audits before and after, run in ChatGPT, Perplexity and Copilot for the same prompt set six to eight weeks apart.
• Whether the content of LLM answers about your brand starts to mirror the language you used in llms-full.txt, which is a soft signal that something is reading it.

You will not get a clean lift number. You may notice nothing changes. That’s information too.

Our take in April 2026

Add llms.txt if your team has the bandwidth, your foundations are in good shape and your buyers genuinely use LLM products to research vendors. Skip it if it’s pulling effort away from work with stronger evidence behind it. Either way, do not let the absence or presence of an llms.txt file convince you that your AI search problem is solved or unfixable.

We expect the picture to clarify over the next twelve to eighteen months. If one of the major engines publicly confirms it uses llms.txt as a meaningful signal, the calculus shifts. Until then, treat it as a tidy-up step rather than a strategy.

If you’re trying to get visible in LLM answers and not sure where to start, tell us about your business. We can usually point teams towards the highest-impact starting place inside our AI SEO work.