How AI decides who to recommend

AI doesn’t recommend whoever spends the most — it recommends whoever the sources it trusts agree on. It builds its answer from third-party signals: reviews, directories, expert mentions, and press — not from your ad budget, and not even mainly from your own website.

That’s why being cited isn’t the same as being recommended. A June 2026 study found a brand’s own page was cited as a source while the AI recommended its competitors about 69% of the time.¹ Your site can help the AI find you; what makes it choose you is the agreement of sources beyond your control.

Reviews alone don’t carry it either: in our own research, a business with 587 reviews was still invisible in most answers.²

The lever that moves AI is third-party consensus you earn — accurate listings, genuine reviews, and real mentions in the places AI reads.

That’s also the part you can’t fake your way into, and the reason earned visibility lasts: it’s built on signals the AI already trusts.

The three signal families

Across every audit we’ve run, the businesses that got recommended were strong on the same three fronts. They work together: strength in one rarely compensates for absence in another.

1. Consensus. Do the sources the AI reads agree about who you are and that you’re good? This is reviews, directory ratings, “best of” lists and real mentions — and it’s the heaviest of the three.
2. Certainty. Can the AI pin down what you do, where, and for whom? Inconsistent names, addresses, hours and service lists across the web make an assistant hedge, and a hedging assistant recommends someone else.
3. Quotability. Have you published a clear answer it can lift and attribute? This is where your own site earns its keep — not by praising you, but by answering the question.

Why your own website still matters

It would be easy to read the 69% finding as “your site is irrelevant.” That’s the wrong lesson. Your site is what makes you legible — it tells the assistant what you do, in language it can quote. What it can’t do on its own is make you credible, because assistants discount self-description by design. Credibility comes from outside.

So the site is necessary and insufficient at the same time. Without it you can’t be cited; with only it, you get cited while someone else gets recommended — which is exactly the pattern the study measured.

What you can’t shortcut

The uncomfortable part of this model is that the heaviest signal is the one you control least. You can write your own pages this afternoon. You cannot write the directories’ view of you, and attempting to — bought listings, fabricated reviews — tends to introduce exactly the inconsistency that makes assistants less certain, on top of breaching platform rules.

What you can do is earn it deliberately: get listed accurately everywhere that matters in your field, ask real customers for reviews properly and consistently, and publish genuinely useful answers to the questions asked before “best.” That is slow, and it compounds.

One caveat worth keeping in view: none of this is static. Each engine weighs these signals differently — see does each AI engine recommend differently? — and the answers shift as engines update. That’s why we re-run the same questions every month rather than treating one audit as the finding. We keep our own record, wins and losses, at our visibility, in public.