Original research
AI does not treat every industry the same way.
We put hiring questions to ChatGPT, Gemini, and Perplexity in the same
five metros for each industry below, and saved every answer. All three
will name businesses. What changes by trade is where they get the
answer from — and that determines where the work starts.
36.4%
of sources cited about law firms
were directories, not company websites
20.0%
of sources cited about roofing companies
were directories, not company websites
21.3%
of sources cited about hvac companies
were directories, not company websites
15.2%
of sources cited about dental practices
were directories, not company websites
10.7%
of sources cited about med spas
were directories, not company websites
The spread matters more than any single number. In a trade where
assistants lean on directories, the work is getting your listings
complete and consistent, because that is the evidence being read. In a
trade where they lean on companies' own sites, the work is on the site
you control — which is the better position to be in, and the
rarer one. Each industry page below breaks down its own result,
including who got named and which sources decided it.
Caution tracks regulation, not difficulty
One pattern held across every trade we measured. Assistants answer
freely about work you hire for your house, and hedge about work you
hire for your body or your legal position. The share of answers that
named nobody, worst assistant per industry:
If you are at the cautious end of that list, the assistant is not
withholding your name because your competitor outranked you. It is
declining to make a recommendation at all, and the way through is
evidence it can safely cite — credentials, licensing, and
verifiable third-party sources — rather than louder marketing.
Method. Identical prompts issued to each assistant's
public API across Las Vegas, Tampa, Denver, Phoenix, and Austin, split
between commercial phrasing and the situational phrasing a buyer with
the actual problem uses. Every assistant was queried with web search
enabled, because that is what a buyer's app does; without it these
models answer from memory and decline to name almost anyone, which
measures the model rather than the market. Business names were resolved
across spelling variants within each metro before counting. Assistants are
non-deterministic, so these figures describe one collection run rather
than a permanent state of the world. Want the same measurement for your
own business? The
free AI visibility report
runs it against your market.