Getting Found by AI: Structured Data and llms.txt
A growing share of "where should we eat" now happens inside an assistant rather than a search box. The guest asks a question, gets three recommendations with hours and a price range, and never opens a website at all.
You cannot opt out of that. You can decide whether the summary is accurate.
The difference between reading and rendering
A person reads your site by looking at it. A machine reads it by parsing it. Those go wrong in different ways.
Your opening hours might be beautifully set in a styled block that a human understands instantly. To a parser, that block is a string of text in a container with no meaning attached. It does not know that "Tue-Sat" is a range of days, or that the numbers next to it are times.
Structured data fixes that by attaching meaning explicitly. Marking a block as opening hours, with named days and machine-readable times, means the system does not have to guess.
What to mark up
For a hospitality business, the useful set is small:
- The business itself — name, address, phone, hours, price range, languages, service area
- The menu, if you have one online
- Frequently asked questions, where the page really is questions and answers
- Articles, for anything you publish
- Breadcrumbs, so the site's shape is legible
Two rules matter more than completeness. Mark up only what is genuinely on the page — inventing structured data for content that does not exist is penalised. And describe your business as one entity that every page points back to, rather than repeating slightly different versions of your details on each page. Contradictory copies are worse than one clear record.
llms.txt
A newer convention, borrowed in spirit from robots.txt. A plain text file at the root of your site that tells an assistant what the site contains, in the order a human would explain it: what the business is, the key facts, the main pages with one-line summaries, and where to find the rest.
It exists because crawling a modern website is expensive and lossy. A site rendered by JavaScript, wrapped in navigation and cookie banners, is a poor way to learn that a café opens at eight and does not take bookings. A short text file is a good way.
It is not a ranking mechanism and nobody guarantees it will be read. It costs very little, and it removes the excuse for a wrong answer.
The part people get wrong
Structured data is not a way to claim things. It is a way to state things you can prove on the page.
Marking up reviews you do not have, or a price range you do not honour, produces exactly the outcome you would expect — a manual penalty, or an assistant confidently telling a guest something untrue about your business. The second is worse, because you will never see it happen.
Where to start
Get the basics right first: name, address, phone and hours, identical everywhere, marked up once, correct through the holidays. That single record is what nearly every automated answer about your business is built from.
Everything else — the menus, the articles, the text file — is refinement on top of it.
