← All posts

How AI assistants decide which sites to recommend

When someone asks an AI assistant to "find a good waterproof jacket" or "book a plumber near me", the assistant does not return ten blue links. It picks. Often it recommends one, two or three options, and acts on them. Being in that shortlist is the new front page, and how you get there is not a mystery.

It can only recommend what it can read

An assistant cannot recommend a site it could not understand. First it has to discover your site, read what you sell or do, and trust the information is current. If your content is locked inside markup, pop-ups and scripts that only render for a human, the assistant has less to go on, and reaches for a competitor it can read cleanly. We covered that baseline in Is your website invisible to AI agents?; recommendation is what is at stake when you fix it.

The signals that tip a recommendation

Among sites an assistant can read, it favours the ones that make its job easy and low-risk:

  • Clear, machine-readable descriptions of what you offer, so it can match you to the request with confidence.
  • A curated guide to your best pages (an llms.txt), so it does not have to infer your structure.
  • Clean content on request: Markdown instead of a page it has to strip down.
  • Real, described capabilities: a way to check stock, get a quote or start a booking, rather than a form it has to reverse-engineer.

These are the same seven signals agents check for readiness, and they double as recommendation criteria. An assistant choosing between three shops reaches for the one that published a clean description and a working way to act.

Trust and freshness do the tie-breaking

Between two readable sites, an assistant leans on confidence. Accurate, consistent, up-to-date information wins; contradictions and stale pages lose. Inventing capabilities you do not have backfires. If an assistant tries an action you advertised and it fails, you are the site it stops trusting. Honesty is not just ethics here; it is ranking.

What this means in practice

You do not earn a recommendation by gaming a score. You earn it by being the easiest, most trustworthy site for an assistant to read and act on. We took a real shop through exactly that. See how we took a shop from 6 to 96.

See how an assistant reads your site today. Run a free AgentVisible scan. No sign-up, just a URL.

See where your site stands

Run a free AgentVisible scan. No sign-up, just a URL.