Data · get my business recommended by ai
We asked an AI assistant to recommend a supplier in 34 Ontario cities. Your own website decided the answer 29% of the time.
An original study of 136 AI recommendation queries across 34 Ontario cities. Seven in ten citations came from a page the business does not own.
Lasse Pettersen
I ran 136 buyer questions past an AI assistant with web search turned on, one for each of 34 Ontario cities across four service categories, and recorded every business it named and every page it cited. 2,064 business listings and 760 citations later, the result is not the one I expected.
70.8% of the citations pointed at a page the recommended business does not own. In 51 of the 134 answers that carried any citation at all, not one source was the website of any business being recommended.
That is the headline, but it is not the useful part. The useful part is that the number moves enormously by category, and the direction it moves in tells you which fix is worth paying for.
What the study did
One question per city per category, phrased the way a buyer phrases it, held identical across all 34 municipalities so that location was the only variable. “Who is the best plumber in Barrie, Ontario? Name specific companies.” The same question for Ottawa, Thunder Bay, Chatham-Kent, Sault Ste. Marie, Collingwood and 28 others.
Four categories, chosen to span how heavy the directory layer is in each: plumbing, commercial cleaning, bookkeeping, and machining and welding. For every one of those 136 questions I also pulled the Google Maps ranking for the equivalent search, so the two answers could be compared.
Nothing here is a survey and nothing is an opinion. Every figure below comes out of a stored response, and the whole set can be re-run.
Where the AI citations actually come from
| Where the citation pointed | Share of 760 citations |
|---|---|
| A third party page | 70.8% |
| The recommended business’s own website | 29.2% |
126 distinct third party domains appeared. That sounds like a wide field until you look at the concentration.
The five domains carrying nearly half the third party citations
| Domain | Share of third party citations |
|---|---|
| threebestrated.ca | 18.2% |
| proscore.ca | 12.5% |
| bbb.org | 6.1% |
| bestprosintown.com | 5.2% |
| yellowpages.ca | 4.6% |
Those five supplied 46.7% of every third party citation in the study. threebestrated.ca appeared in 45 of the 136 queries on its own, and proscore.ca in 21.
Most Ontario owners I speak to have heard of two of those five. The other three are deciding answers about their business in their own city, today, and nobody in the building has ever opened the page.
The deciding layer changes with the category
This is the part worth acting on. Split the same 760 citations by category and the picture inverts.
| Category | Citations from the business’s own site |
|---|---|
| Plumbing | 4.0% |
| Commercial cleaning | 26.9% |
| Bookkeeping | 45.7% |
| Machining and welding | 49.4% |
A plumbing company’s website is very nearly irrelevant to whether an assistant recommends it. A machine shop’s website is close to the only thing that matters.
Why the two ends behave differently
The categories with a thick directory layer are the ones somebody already built a business around. Home services have been aggregated and re-aggregated for twenty years, so when an assistant goes looking for “best plumber in Oshawa” it finds a dozen pages that have already done the ranking and it uses them.
Nobody built a national roundup of Ontario machine shops. So the assistant reads the shops’ own pages, and the shop that describes its tolerances, its materials and its lead times in plain sentences is the one that gets named.
Where your own website still wins the AI answer
The pattern holds in a second place too. In the five Ontario cities over 500,000 people, own-site citations ran at 16.8%. In the 19 cities under 150,000 they ran at 30.1%.
The practical version of that for a smaller market
If you sell in Peterborough, Sarnia, Brantford or Collingwood, your own service pages are close to twice as likely to be the source an assistant quotes as they would be in Toronto or Mississauga. The smaller your market, the more the words on your own site are the asset.
The trades version in a large city
If you run a plumbing or a cleaning business in the Greater Toronto Area, the opposite is true and rewriting your homepage will not move this. Your presence on the five domains above will.
Your Google review count barely enters into it
Every answer in this study arrived with a block of local businesses and their Google ratings, which makes the comparison direct: was the business the assistant named first also the most-reviewed business it had in front of it?
It was, in 14.2% of queries.
| Measure | Median across 136 queries |
|---|---|
| Google review count of the business named first | 32.5 |
| Google review count of the most-reviewed business in the same answer | 197 |
The extreme cases are not close. In Oshawa the assistant’s first pick was a plumbing firm with a Google review count of 2, ahead of a national brand in the same list with 7,760. In Waterloo it named a firm with a review count of 1 ahead of one with 2,787.
This is the single most counterintuitive number in the study, and it is worth being careful about what it means. It does not mean ratings are worthless. It means the assistant is not ranking by them. It is ranking by which page describes the work in the most specific prose, and a small firm that explains what it actually does can beat a large one that does not.
The AI answer is not the map pack
| Comparison | Result |
|---|---|
| Average overlap, AI top three against Maps top three | 1.13 of 3 |
| Queries where the two top threes shared nobody | 25.7% |
| AI top three businesses absent from the Maps top 20 entirely | 16.2% |
A quarter of the time, the business winning the AI answer and the business winning the map pack are two disjoint sets. In Brampton the assistant’s top three bookkeeping firms were Gondaliya CPA, Profit Pioneers and Think Accounting. Google Maps returned Accountor CPA, Trillium CPA and Pace Accounting. No overlap.
If you have been tracking your map pack position and treating it as your local visibility number, it is measuring one of two channels.
Municipal boundaries leak
27.9% of the 2,064 businesses named for a city had an address outside it. Ask about Barrie and you get firms in Innisfil, Minesing and Orillia. Ask about Chatham-Kent and you get Comber and Dresden.
Some of this is ordinary service-area overlap and should not be read as an error. The useful reading is the other direction: a firm one town over can appear in a city answer it never targeted, which is a cheaper opportunity than most local marketing.
What this study does not show
Three limits, stated plainly, because a study that only reports its wins is a sales document.
One assistant, one week, one country setting. Answers vary between runs and between products, and this is a snapshot rather than a trend until it is run again.
45.6% of the businesses named came without a link. That looked at first like a large population of businesses with no website. It is not. Cross-checking against the Google Maps record for the same businesses, only 66 of 2,064 had no website anywhere, which is 3.2%. The missing links are the assistant’s formatting, not the businesses’ absence.
And four categories are four categories. The gradient from plumbing to machining is clear across 34 cities, but it does not tell you where veterinary practices or freight brokers sit. It tells you to go and measure your own.
What an Ontario business should do with this
Four steps, in order, and the first two cost nothing.
Ask the question yourself, in your own city, and read what gets cited. Not “is my business online” but “who is the best X in Y, Ontario”. Then open every source the answer used. That list is your actual competitive set for this channel.
Count the sources that are your own site. If it is zero, your website is not the lever in your category and no amount of rewriting it will change the outcome. Fix your presence on the pages that were cited instead.
If your own site is being cited, make it specific. The businesses winning these answers with small review counts are winning on described detail: materials, lead times, service areas, what they do not do. Vague pages lose to specific ones here far more decisively than they do in ordinary search.
Then check whether any of this is your real problem. Statistics Canada put Canadian business AI adoption at 19.2% in the second quarter of 2026, and the leading reason non-adopters gave, from 40.0% of them, was that AI is not relevant to what they do. For most businesses the larger and duller opportunity is not being recommended by an assistant. It is the 30 hours a month somebody spends retyping the same order into two systems, which is what a workflow automation build is for, and what the $999 assessment exists to find and cost.
Being invisible to an AI assistant is a real problem worth an afternoon. It is rarely the most expensive one in the building. For the trades version of that arithmetic, see the four automations that pay in a trades business, and for the general version, seven jobs AI does well in a small business.
How this was measured, so you can check it
136 questions, 34 Ontario cities, 4 categories, one question per city per category, collected on August 26 and 27, 2026 through the DataForSEO ChatGPT scraper with web search forced on and the country set to Canada. Google Maps rankings for the equivalent query were collected in the same run at a depth of 20 results.
A citation counted as the business’s own when its domain matched a domain the answer itself attached to a business it named. Businesses were matched between the two data sets on a normalised name comparison, with the corporate suffixes stripped, so that one company spelled two ways counted once.
Two of the 136 answers carried no citations and are excluded from the citation percentages. They are included everywhere else.
The collection script, the analysis script and all 272 stored responses are in the site repository. Every number above re-derives from them.