2026-08-03 · Jack Stovell

How to get your business cited by AI assistants

More buying research now starts as a question to an AI assistant: "best review-request tool for small businesses", "who makes X in the UK". The assistant searches, reads what it can, and answers with the companies whose pages survived that reading. For a small business this is a rare open window — the incumbents' advantage in domain authority matters less to a system that reads pages than it did to one that counted links. The end state of the layers below is an AI-readable profile for the business — what that is, and what it cannot guarantee, is defined on its own page.

We run several small products ourselves, publish our own AI-visibility benchmark, and got every one of our businesses AI-readable the hard way. This is the layer-by-layer version of what that took.

Layer 0: can crawlers read your site at all?

Before anything clever: several major hosts and CDNs now ship with AI crawlers blocked by default — a switch you never flipped, silently returning 403s to exactly the systems you want reading you. We've hit this on our own domains more than once. Fetch your site with an AI crawler's user agent, or run it through our free checker, before spending an hour on anything below.

Layer 1: facts a model can repeat

An assistant recommending your business needs to state what you do, for whom, and at what price — without inferring it from marketing copy. Audit your homepage for the sentence a stranger would need: most say "transform your workflow" somewhere prominent and never plainly say "we make review-request software for UK service businesses, from £6.99/month". The plain sentence is the quotable one.

Layer 2: identity markup that says Organization

Schema.org Organization JSON-LD tells systems your business is a company — its name, its own domain as the canonical url, its founder, its other surfaces as sameAs. One modelling rule matters more than the rest: your company's url must be its own website. Directory pages, social profiles and everything else belong in sameAs. Markup that gets this backwards teaches models that some platform's page IS your business.

A founder link matters too: "who is behind X?" is one of the questions assistants actually get asked, and it's answerable only if some page connects the company to a person a model can resolve.

Layer 3: an llms.txt and a profile built to be read

The emerging convention for AI-facing sites is a plain-text summary at /llms.txt — the facts, without the design. Every one of our products publishes one, and they're what we use as the source when we publish anything else about those businesses, so no surface ever says something the company's own site doesn't.

A hilyt business profile packages this whole stack: Organization JSON-LD with your domain as canonical, founder resolution, dated claims each carrying a source URL and a one-click owner confirmation (your own tick that the wording is right — labelled as such, not passed off as independent verification), crawler-rendered HTML, and a plain-text llm.txt card. It's free, and the same page also works as the "who are we" link humans ask for.

What we won't tell you

We won't tell you this guarantees recommendations — model output isn't for sale, and our own AI visibility benchmark shows exactly how far along we are on our own prompts, misses included. What the work above changes is the denominator: when the search runs, your business is one of the ones the assistant could cite. Most of your competitors currently aren't.

See your business profile drafted in 30 seconds — free