Has AI search changed how Canadians buy, and the evidence that exists so far

Something has changed in Canada and it is specific rather than sweeping. On 22 October 2026, across 64 blind readings of the ChatGPT and Gemini product surfaces on Canadian buying questions, the engines produced 2,992 citation URLs across 974 distinct hosts, named a supplier in 31 readings and named nobody at all in 33. They named six Canadian agencies unprompted, 21 namings in total, alongside 48 namings of software products. The change a Canadian business should act on is not that buyers left Google. It is that a second answer exists, it is generic in most categories, Canadian firms do appear in it, and the two engines behave so differently that treating them as one channel would mislead you.

What was measured, and the three limits

The bank was 198 Canadian buying questions in 33 batches, put to both engines through a data provider. Canada was named inside the text of every question, because the provider accepts gl, hl and location parameters and ignores all three, so the prompt is the only market lock that holds. Batch 33 names us in the question and therefore measures a ceiling rather than a market, so it is excluded from every figure. That leaves 64 blind readings, 32 per engine.

Three limits, stated because they change how the numbers should be used. This measures what the engines answer, not how many Canadians ask. It covers two of the five commercial surfaces, so Perplexity, Claude and Google AI Overviews sit outside it. And it is one day, which makes it a baseline rather than a trend.

The engine split, which is the Canadian finding

Reading	ChatGPT	Gemini
Blind Canadian readings	32	32
Ran a live web search	32 of 32	20 of 32
Named a supplier that was not us	9	17
Named no supplier at all	23	15

Gemini named a rival nearly twice as often as ChatGPT on Canadian questions, while ChatGPT read the live web far more often. That is the opposite balance to what the same bank produced for another market we ran on the same day, which is why a national average across markets would be useless and why a blended score across engines would be worse. Report both engines, every time.

What has moved, in three sentences

• A second answer exists and it is usually generic. Thirty three of 64 Canadian readings named no supplier, so the engines answered the buying question with a framework.
• Canadian firms are visible in it, thinly. Six agencies named 21 times: Major Tom in 8 readings, Konstruct in 6, Jelly Digital, Seota and Pilot Digital in 2 each, Edkent once. Software took 48 namings, led by Semrush at 16, HubSpot at 12 and Ahrefs at 9.
• The sources are Canadian and official. Four federal properties appeared in 36 readings between them: canada.ca in 10, competition-bureau.canada.ca in 9, ised-isde.canada.ca in 9 and www150.statcan.gc.ca in 8. City specific domains appeared high too, torontoseo.com in 13 and ottawaseo.com in 7.

One entry in the raw host data, ccn.web.tr, is the data provider's own fetch infrastructure rather than a publisher and is excluded from every source analysis.

The three decisions to make on this evidence

• Stop reporting it as one channel. On this reading the two engines differ on live search behaviour and on willingness to name a supplier, and a single figure hides both.
• Work the empty answers first. A reading that named nobody has no incumbent to displace, so the cost of taking it is publishing a specific, dated answer. Sort your own question list by that before spending anything.
• Publish toward Canadian official sources. The engines leaned on federal data properties and regulator positions. A page built on Statistics Canada data with a date, or one that states the Competition Act position accurately, agrees with a source the engine already trusts.

What this does not show

We have no Canadian share figure for how many buyers consult an assistant before contacting a supplier, and we will not publish one we did not measure. The surveys in circulation differ on whether they asked about shopping, research or supplier selection, on whether they sampled consumers or business buyers, and on whether an AI answer inside a search results page counted. Those three choices move such a figure by tens of points. What is honestly available to a Canadian business is its own reading, taken in an afternoon, specific to the questions its own buyers ask.

How to take your own Canadian baseline

• Write twelve buying questions in the words your Canadian customers use, naming Canada or the province in the text of each.
• Run them in French as well if you sell into Quebec, as a separate question set rather than a translation.
• Put each to both engines, save the answer text and the source list, and record the date and the engine.
• Score three ways. Named you, named a rival, named nobody, and count the hosts.
• Re-read in thirty days on the same questions and report the engines in separate columns.

If the French question set is the part you have not planned, the French first position in Quebec and the privacy law touching your tracking are set out in French first by law, and Law 25 on your tracking. For the comparison of what each queue is better at before you reallocate anything, see what each is better at and what the data shows.

Questions about AI search and Canadian buying behaviour

Are Canadians using assistants instead of Google?

Our evidence speaks to what the engines answer rather than to how many people ask. A second answer now exists and it is generic in most Canadian categories, which is the part a business can act on.

Which engine matters more in Canada?

They matter differently. Gemini named rivals nearly twice as often and ChatGPT read the live web nearly twice as often, so recency and naming behave oppositely across the two.

Why were Canadian agencies named when software dominates?

Because enough attributable published material exists about them for an engine to use. Twenty one namings across six firms is thin and it is not zero, and it sets a reachable standard.

Is 64 readings enough to act on?

Enough for the three method decisions above, which are about reporting and sequencing. Not enough to publish a national trend, which is why we have not.

How often should this be re-measured in Canada?

Monthly, holding the questions, the engines and the denominator constant, with French counted separately if it applies.

Why the second answer deserves its own Canadian reporting line

A Canadian marketing report that covers search and media covers one queue well and the other not at all. Ours found 64 Canadian readings in which we were named zero times, a rival 31 times and nobody 33 times, with the two engines disagreeing sharply. AI Knows Us reads ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews with Canada named inside the question, reports each engine separately with its denominator, and keeps the raw answer text so the line has something behind it. Outside the United States and the United Kingdom the cheapest plan that reads all five engines is Growth at 199 dollars a month with 100 buyer questions watched, read on aiknowsus.com/pricing and aiknowsus.com/llms.txt on 22 October 2026.