How do I get my business mentioned by AI assistants?

The short answer: publish things worth quoting, and get findable in ordinary search first. Speed makes almost no difference.

Ask an assistant to recommend an accountant in Gloucester and it names three or four firms. There are far more than four accountants in Gloucester, so something decides which get said out loud.

We test this for every business in the Local Digital Visibility Index: five natural-language questions per sector, the kind a customer would actually type. Across 1371 businesses, 355 were named at least once and 1016 never were — 74% are absent from every answer.

Every figure here is computed from the published data when this page is built.

What separates the named from the unnamed

Measured on	Named by the assistant	Never named	Difference
Speed & CWV	61	64	-3
Technical	88	75	+13
Local presence	83	73	+10
Visibility	37	0	+37
Content & trust	55	41	+14

Speed & CWV differs by -3 points. On a 0–100 scale across 1371 businesses, that is nothing. The firms an assistant recommends load at almost exactly the same speed as the firms it ignores.

Visibility differs by 37 points, the largest gap of anything we measure.

Ranking first, then being quotable

The relationship between ordinary search and AI answers is one-directional, and the asymmetry is the practical part:

• 411 businesses rank in ordinary search and are still never named.
• 31 are named while ranking for nothing — out of 1371.

So appearing in ordinary results is close to a precondition — almost nothing gets recommended without it, and 31 exceptions in 1371 is thin enough to be noise rather than a route you could take deliberately. But it is nowhere near sufficient: 411 businesses have cleared that bar and are still not chosen.

Being on page one buys a ticket to be considered. Most businesses holding that ticket are still not picked.

What to actually do

In the order the evidence supports:

1. Answer real questions in your own words, on your own site. An assistant assembles an answer from sources it can find and attribute. A site with three pages of service descriptions offers nothing to attribute, however fast it loads.
2. Make the basic facts explicit and machine-readable — who you are, what you are accredited to do, where you work, what you charge if you can say. Structured data helps once the facts exist; it cannot invent them.
3. Fix ordinary findability first, since it is close to a precondition. Our answer on why businesses do not show up covers that.
4. Do not buy a “GEO” package that is a speed audit in new packaging. The data says speed differs by -3 points between named and unnamed.

What this does not tell you

One assistant. Every figure comes from Perplexity, on five prompts per sector. It is not evidence about Gemini, ChatGPT or AI Overviews, and our methodology page was corrected for once implying otherwise. A business absent here may be named elsewhere.

Five prompts is a small basket. Ask differently and a different set of names comes back. That instability is a real property of the thing, not noise we failed to remove.

Correlation, not instruction. Businesses that get named also tend to rank and to publish. This data cannot tell you which of those causes which.