“Find me a good mortgage broker” is now a prompt, not just a search. When an assistant answers it, a selection happened — and unlike a rankings page, you can’t see the leaderboard. After a year of testing these answers across Canadian markets, here’s our working model of how the selection goes.
The three gates
Gate 1: Can the model verify you exist? Engines lean on consistent, cross-confirmable facts: your name, license, brokerage, and market appearing identically across your site, Google profile, industry directories, and licensing registries. Contradictions (old brokerage names, dead profiles) read as risk, and assistants are risk-averse about recommending humans.
Gate 2: Is there third-party evidence? Reviews, media mentions, and being cited or linked by other sites in the niche. The engines behave like a cautious referral partner: they want someone else to have vouched first.
Gate 3: Have you answered the user’s actual question before? This is the strongest lever you directly control. Assistants prefer recommending a source whose content already answers the user’s situation — the broker with the definitive page on, say, buying after consumer proposal will get named in exactly those chats.
The strategic consequence
Niching stopped being optional. “Great all-around broker” is unciteable; “the person who wrote the best Canadian guide to X” is a retrieval target. Pick the files you love doing, own their question space completely, and let the assistants learn that pattern. The generalists will keep wondering where the calls went.