“Make good content and the AI will recommend you” is not useful advice. We went at it from the other end: we traced back the answers of brands that are already being recommended and checked, citation by citation, what the engines said they were reasoning from. Four patterns came out of it.
Pattern 1 — a dedicated page that maps onto the question
Engines cite pages, not sites. A question about patellar luxation surgery pulls the clinic’s dedicated patella clinic page; a question about division of assets in a divorce pulls the firm’s asset division page. The most extreme version in our data was law firms that put the question keyword in the domain itself — running a separate site for divorce work — and the engine cited “operates a site specialising in divorce” as its reason for recommending them. One question, one page. That is the base unit.
Pattern 2 — publish specific numbers about yourself
Share of measured citations that point at the business’s own website — the single largest supplier of an AI’s stated reasoning is the business itself
One vet clinic’s line, “over 10,000 orthopaedic surgeries”, was lifted word for wordas the stated reason by two different engines. Another clinic’s own case write-up — a grade 2 patella in a 3.9 kg Pomeranian — became the recommendation rationale. An AI is not a verification body: it treats a specific number written on your own site as credible evidence. Surgery counts, years of practice, equipment you own, written as figures, get cited. “The best care” does not.
Pattern 3 — structured platform profiles
Vertical platforms — Lawtalk for lawyers, JobPlanet for employers, Diningcode for restaurants, Fitpet for pets — are read repeatedly when an engine needs to compare several businesses. In our data, JobPlanet ratings and review counts were quoted directly as the reason for a recommendation more than once. Filling in the blanks on those profiles and keeping them current is the highest return-per-hour work in this whole list.
Pattern 4 — get your name into third-party content
“5 recommended X” listicles are the format engines find easiest to lift when assembling an answer. We watched one brand enter a recommendation list after appearing in a single Tistory post. Which platform carries you determines which engine reaches you — that mapping is in the engine playbook.
The uncomfortable part — and the gap it leaves
You will have noticed the problem. Unverified self-description is being laundered through an AI and distributed as a “recommendation”. The structure rewards exaggeration. The other reading is that if you happen to have real, specific numbers, this is the cheapest moment there will ever be to claim the ground.
And there is ground to claim: our data still contains questions where not a single brand was named by name — including some of the highest-volume ones, like supplementary medical insurance. Which questions are still empty is listed under the open-spots section of statistics.