Open our own llms.txt in a new tab. Forty-five lines of plain markdown describing what Built Local is, what each major page is for, and where the canonical content lives. Total size: under 2KB. Every site we build ships with one.
This is why.
What llms.txt actually is
llms.txt is a community-proposed file at the site root (like robots.txt) that gives AI crawlers a hand-curated map of the site. The format is flat markdown with H1, an optional blockquote summary, and H2-grouped link lists. The spec is at llmstxt.org.
It is not a magic file. It does not force any crawler to do anything. What it does is reduce the cost for an AI crawler to understand the site - skip parsing the JS shell, skip guessing at the IA, just read the structured summary. When the cost is lower, the crawler is more likely to do the work.
What we put in it
For a typical Built Local trade site, the llms.txt has:
- H1: the business name.
- Blockquote: one-line summary ("Gas Safe plumber covering Newcastle, Gateshead and North Tyneside since 2014").
- A paragraph or two of context on services and areas.
- "## Pages": every public landing page - home, services, suburbs, contact, about, FAQs.
- "## Services": each service page with a description.
- "## Areas": each suburb/town page.
- "## Optional": low-priority pages (privacy, terms).
What we deliberately exclude: the admin dashboard, the login flow, the booking widget endpoints, anything user-specific, anything behind auth. Same rules we'd apply to a sitemap.
The Perplexity referral data
We instrumented Perplexity referral traffic on six clients before and after adding llms.txt, with no other changes in the same window. Results after 90 days:
- Perplexity referrals up by an average of 38% (small absolute numbers, but a real and consistent lift).
- Time on site from Perplexity referrals up by 19% - we suspect the LLM is sending the user to the more correct landing page because it now has a structured map.
- No measurable change in Google traffic - the file is for AI crawlers, not Google.
For ChatGPT it's harder to attribute because ChatGPT's referrer is sparse, but our qualitative read of citation patterns is that llms.txt-equipped sites are cited more often for the queries they're actually best-positioned to answer.
The generator
We don't write llms.txt by hand. Every Built Local site has a generator that pulls from the same data source as the sitemap (resources, services, locations) and produces the file at build time. It enforces the exclusion list, formats correctly, and keeps the file under 5KB. The whole thing is about 80 lines of code.
The reason the generator matters: a stale llms.txt is worse than no llms.txt. If you list a page that no longer exists, the crawler reads a 404 and learns nothing. The generator makes it impossible to forget.
What about the cynical view?
"It's a community proposal, not a spec. The big LLMs haven't formally committed to using it. You're shipping nothing."
Possibly. Even so, the upside is asymmetric. Cost: 2KB of static markdown and 80 lines of generator code, both of which we wrote once and reuse. Downside if AI search never adopts it: zero - the file harms nothing. Upside if even one major LLM keeps using it: a measurable referral lift that compounds. The expected value of shipping is positive even under pessimistic assumptions.
Why this matters for UK local web design
Local businesses are dramatically under-represented in AI search recommendations right now. The recommendation engines lean on big-brand signals because the long tail is poorly structured. A 2KB file is a cheap way to make a local trade legible to those engines, before competitors catch up.
Every Built Local site ships with the generator and the file. No client has ever asked us to remove it. See the AI citation field note for the wider strategy, or browse the AI search pillar for the long version.