Our research
What 61,848 AI citations say about local AI search.
Everyone in this category sells a dashboard. Nobody publishes their data. This is ours: what the AI engines actually read and cite when someone asks for a local business, measured across 941 UK businesses. It changed how we work, and it will probably change what you believe about AI visibility.
What we did
We audited 941 UK local businesses across trades: plumbers, locksmiths, accountants, electricians, valeters and more. For each one we asked ChatGPT and Gemini (plus Google's AI Overview where Google showed one) the questions their customers actually ask, recorded which businesses got named, and logged every source the engines cited. 61,848 individual citations, each tied to a trade and a town. These are our own measurements, not a vendor's whitepaper.
Finding 01
What AI reads is trade-specific.
One directory was cited in 394 of our 411 plumber audits. In our accountant audits it appeared in one of 80. Same engines, same country, same month. The sources AI trusts for one trade mean nothing for another, which is why a generic "get listed everywhere" checklist cannot work: half of it is effort spent where your trade's answers are never sourced from. So the first thing we do is measure your market, not run a checklist: which sources AI actually cites for your trade, in your town.
Finding 02
Being listed is not being named.
We measured one business 447 times before doing any work for it. It already sat on the exact page AI cites most for its trade. It was named zero times. Presence on a cited source gets you read; it does not get you recommended. The difference between the two is where the actual work is, and it is invisible unless you measure. So the work is making you quotable, not just present: pages that say specifically what you do and where, in the words people ask with.
Finding 03
The levers people assume mostly aren't.
Structured data (schema markup) is the standard advice for machine readability. In our measurements it made no measurable difference to being named: 35% of named businesses had it, against 32% of everyone else. Website quality surprised us more: many of the businesses AI names have worse websites than businesses that never get named at all. We still fix both, as hygiene. But the lever is being present and consistent in the sources AI reads for your trade, not polish. So that is where we spend the effort: pages AI can quote and details it can trust, rather than a prettier site.
The wider picture
Our findings line up with the largest study published in this space: Yext's analysis of 6.8 million AI citations for local queries found 86% came from sources the business already controls. Its own website, its profile, its listings. The answer to AI visibility is mostly work on things you own, which is bad news for anyone selling you a dashboard and good news for anyone willing to do the work.
Source: Yext, 6.8 million AI citations from location-specific queries across ChatGPT, Gemini and Perplexity, July–August 2025. First-party websites 44% plus listings 42%. Yext sells listings management, so this is vendor-interested research; it is used here as a supporting figure only, never a headline one.
What we do about it.
The findings above are the shape of the work. A free check that shows what AI says about you today. Then a four-week project: your details made consistent everywhere AI reads, and service and area pages written the way people ask. At week four, the original questions are re-run on the same engines and the before-and-after reports are compared side by side. After that, the monthly, because the engines re-rank constantly; it starts once your claim window closes. It starts with the check, which is free and run by hand.
Get my free AI check