The five things AI checks before it recommends you

Ask ChatGPT for the best personal injury lawyer in your city, or the best orthodontist, or the best agency — and it names someone. Not a list of ten blue links. A name. The obvious question for every business owner watching this happen: how does it choose?

It isn't ranking. It's believing.

A search engine orders documents. A language model composes an answer — and to do that, it has to decide who it believes. Those are different jobs, and they check different things. After running audit prompts across the major engines for months, the pattern behind the recommendation is consistent. Five checks, roughly in order:

1. Can it read you?

Before AI can recommend a business, it has to be able to parse who you are, what you do, and where you do it — from your site, your listings, your structured data. Businesses that are perfectly legible to humans are often half-invisible to a machine. This is the most common failure and the least glamorous fix.

2. Does anyone else agree?

Models weight corroboration heavily. One site saying “we're the best” is a claim; a dozen independent sources describing you consistently — reviews, directories, press, professional listings — is a fact pattern. When AI hands a recommendation to a competitor, the competitor usually didn't out-rank you. They out-corroborated you.

3. Is the story consistent?

Conflicting names, addresses, categories, and descriptions across the web don't just dilute you — they make the machine hedge. We've watched engines describe the wrong company under a brand's name because two similarly-named businesses had tangled fact patterns. Consistency is trust, mechanically.

4. Is it fresh?

Engines that browse lean toward sources that look alive. A brand whose most recent third-party mention is three years old reads as historical, not current. Recency isn't vanity — it's a signal the recommendation is still safe to make.

5. Is it structured to be quoted?

Models lift answers from content that already looks like an answer: clear claims, plain language, questions addressed directly. Pages written to impress read worse to a machine than pages written to inform.

What this means practically

None of these five is exotic, and that's the point: being recommended by AI is mostly the compounding result of unglamorous, verifiable work — done in the right order. It's also why we tell every client the same thing: before you fix anything, measure. Ask the engines what they say about you today, and let the gaps set the sequence.

Want the measurement? That's exactly what our Trust Index audit does — scoped with our team, or instant through Ask Taylor.