What is changing
More people now ask an assistant instead of searching: “who should I call about a leaking hot water tank in Maple Ridge?” The answer is not a list of ten links. It is a short paragraph that names two or three businesses, or none. Google does this inside search too, with AI Overviews at the top of some results and AI Mode as a separate conversational search.
The honest starting point is that nobody outside these companies knows exactly how they choose. What we do have is a growing set of studies on what they cite and how often they name businesses. This guide sticks to those.
They name far fewer businesses than the map
For a small business that is both the problem and the opportunity. Being one of the few names an assistant gives is worth a lot, and most of your competitors are not working on it yet.
What assistants lean on
Across the studies, the same kinds of sources keep coming up. In the 2026 expert survey of local ranking factors, the top factors for AI visibility were third-party “best of” lists, dedicated service pages, prominence on industry sites, unstructured mentions and the authority of review sites. [1]
| Source | Why it matters | Evidence |
|---|---|---|
| “Best of” lists on other sites | Lists of the best businesses in a town are the page type ChatGPT cites most often | Strong [2] |
| Your own service pages | A page that plainly says what you do and where is easy to quote | Strong [1] |
| Review platforms | Google, and now Yelp, whose reviews and listings began appearing inside ChatGPT answers in July 2026 | Strong [5] |
| Consistent listings | The same name, address and phone everywhere makes you one clear business, not three possible ones | Strong [1] |
| Google, Apple and Bing profiles | Google's AI features draw on its own business data. Apple Business feeds Siri and Apple Maps, and Bing Places feeds Copilot | Likely |
“Best of” lists come with a warning. Getting onto a genuine list, such as a local paper's roundup or a supplier's installer directory, helps. Publishing your own “top 10 plumbers in Surrey” with yourself at number one does not, and Google moved against self-promotional lists like that in 2026.
What does not work, whatever you are sold
llms.txt
- A file meant to tell AI crawlers about your site. In a 2026 study, 97% of llms.txt files received zero requests. [4] We do not sell it.
Schema as an AI booster
- Structured data is good housekeeping and we add it to pages. The same 2026 study found no measurable lift in AI citations from it. [4] It is not a lever.
A “first place” in ChatGPT
- There is no stable rank to be first in. Ask the same question twice and the list changes, as the next section shows.
Wikipedia for a small trade
- A three-truck plumbing company is not going to get or keep an encyclopedia entry, and it is not what assistants use for local recommendations.
How to measure it honestly
AI answers are inconsistent. In one 2026 study, the same prompts were run 2,961 times, and the same list of brands came back less than once in a hundred runs. But the share of runs that named a given brand was stable. [6] That gives a sound way to measure: ask the same fixed questions many times and count how often you are named.
- Fixed questions, written the way customers ask, with the town in them. Change them only on purpose, and say so.
- Many runs, not one. Report a share, such as “named in 9 of 60 answers”, never a position.
- Clean conditions: no personal logged-in account, which personalises the answer.
- Named and cited, kept apart. Being recommended and having your website linked as a source are different results.
- Re-baseline when the data changes, such as when an assistant adds a new review partner.
That is how the free AI check and our monthly measurement work. Today they ask ChatGPT, through OpenAI's API, and count what comes back. We add other assistants only once we can measure them the same way.
What to do, in order
- Get the basics consistent. Google profile, Apple Business, Bing Places, Yelp and the core Canadian directories, all with the same details. The listings guide has the list.
- Write a real page for each core service, with common questions answered in visible text: cost ranges if you are willing, timing, permits, rebates. How to keep them useful.
- Keep reviews flowing on the platforms assistants read, from every customer, without filtering. The rules.
- Earn a place on genuine local lists and directories you qualify for: trade associations, manufacturer installer lists, the chamber of commerce, the local paper.
- Look at who gets named instead of you, and which sites the assistant cites. That list is your to-do list.
A fair expectation
Nothing here makes a mention certain. Assistants change their sources, sometimes overnight. What you can do is make your business easy to find, easy to confirm and well reviewed in the places they read, then measure the share of answers that name you and see whether it moves.
Sources
- Whitespark, Local Search Ranking Factors 2026, AI visibility factors, an expert survey published November 2025. whitespark.ca/local-search-ranking-factors
- Ahrefs, analysis of the page types ChatGPT cites (“best X” lists at 43.8%), as reviewed 20 September 2026.
- SOCi, 2026 study of AI recommendations for multi-location brands: share of locations recommended by ChatGPT (1.2%), Gemini (11%), Perplexity (7.4%) and the Google 3-pack (35.9%).
- Ahrefs, 2026 studies of llms.txt requests and of structured data in AI citations.
- Axios, report on the Yelp and OpenAI partnership, 23 July 2026. axios.com
- SparkToro and Gumshoe, 2026 study of AI answer consistency: 2,961 runs of 12 prompts.
Where Merkat does this for you: We ask AI assistants what your customers ask and count how often they name you, then work on what they rely on.
AI search visibility →