AI shopping answers in Canada: retailers and sources in the sample

A Canadian shopper who asks an assistant for the best air fryer might be sent to Canadian stores, or to the same US pages an American would see. The October 2026 Canada run lets us look at the sources behind 297 answers and at how the brands compare with the US for categories that share a name. It is a description of one sample, not a measure of how Canadians shop.

What is in the Canada sample?

Canada has 30 categories with prompts in English, written with Canadian wording and prices in Canadian dollars. In the October 2026 edition, each prompt was asked once on OpenAI's API with web search (gpt-6-luna) and once on Perplexity's Agent API with the Sonar model (perplexity/sonar). That gave 297 answers, of which 150 came from OpenAI and 147 from Perplexity. Each request sets Canada as the search location, a setting OpenAI documents as a way to refine results by geography. The methodology has the rest, and every category is listed in the Canada index.

The Canada categories are the October 2026 edition and are frozen. Each has about 10 answers.

Which stores appear in the source lists?

Each row is a store domain, by our reading of the domain name, among the 30 domains that appeared in the most Canadian answers' source lists, from either engine. Editorial and other sites are in the next section. Engines record sources differently (OpenAI's list holds the pages it linked, Perplexity's the pages it retrieved), and Perplexity's much longer lists probably supply most of the entries. Shares are of Canada's 297 answers.

Store domains among Canada's 30 most frequent domains. October 2026 edition, Canada, 297 answers, OpenAI and Perplexity pooled.

So for Canadian prompts, the most frequent store domains were Canadian ones. Amazon's Canadian domain was in 93 answers' lists and its US domain in 14. A few things are worth keeping in mind.

• Several big names are missing. walmart.ca, costco.ca and target.com are not among Canada's 30 most frequent domains, whose last entry is in 10 answers, so each appears in at most 10.
• Retrieved lists cross markets. Perplexity retrieved amazon.ca in 123 answers across the whole edition, but Canada's pooled count is 93. At least 30 of those 123 retrievals were therefore in US or Mexico answers.
• We cannot split the Canada table by engine. Perplexity's longer lists probably supply most of it.
• The retailer list is hand-made. Seven of the 30 most frequent domains are on it. decathlon.ca looks like a store but is not.
• A listed domain is not an influence. A domain in a source list does not show that it shaped the recommendation.

What else fills the lists?

Stores are a minority of Canada's 30 most frequent domains. 10 of the 30 end in .ca, and the other 20 include US stores and editorial review sites: nytimes.com is in 43 answers (14.5%) and rtings.com in 36 (12.1%).

A few Canadian sites that are not stores also appear: inceptionai.ca in 25 answers (8.4%), madeinca.ca in 17 (5.7%) and globalnews.ca in 12 (4.0%). We checked two of them against the sites. inceptionai.ca, a Canadian site that helps people choose products, says in its footer that it answers to its visitors, not advertisers, though the page does not say how it is funded, and madeinca.ca marks Canadian-owned products and companies. The sample cannot say why they were retrieved, and a domain in a list is not evidence that it was used.

Brand counts look like the other markets. Excluding the alternative-to-a-brand question, Canada's 237 answers recommended a mean of 2.3 brands and 1.7% recommended none. The two engines' recommended-brand lists overlapped by a mean of 45.9% per category (brands both engines recommended divided by brands either recommended, averaged over the 30 categories; not answer-by-answer agreement). Mexico's sample in Spanish is covered in a separate post.

How close are Canadian and US answers for the same category?

There are 24 category names that exist in both the US and Canada runs. For each, we took every brand recommended in any answer in the US category and every brand recommended in any answer in the Canadian one, and divided the brands on both lists by the brands on either list. This uses all five October prompts and both engines, including the alternative-to-a-brand question, which lists more brands than the others.

The mean across the 24 is 54.9%. In 18 of the 24 pairs the lists shared half or more of their combined brands, and in 2 they shared under 30%. Leaving out the alternative-to-a-brand question, as the index's brand counts do, the mean falls to 47.6%, with 11 of the 24 pairs at half or more and 4 under 30%. Each pair is one example resting on about 10 answers per market, not a finding about that category. The US side of 11 of these pairs (mattresses, bed sheets, office chairs, cookware sets, chef knives, carry-on luggage, backpacks, water bottles, yoga mats, dog food and face moisturizer) is the October 2026 edition. Those US category pages have since been updated to a different, larger November sample, so their counts will not match the chart.

The comparison is rough on purpose. The prompts are not identical. The air fryers budget prompt reads "best air fryer under $80" in the US category and "best air fryer under $100 CAD" in the Canada category. The alternative prompts can name different brands. And answers vary from run to run, so some of the gap may be ordinary variation that sampling the same market twice would also show.

What can a Canadian brand take from this?

1. Check the .ca store pages. In this sample, amazon.ca and bestbuy.ca were the most frequent store domains for Canadian prompts. Make sure your product pages there match your own site, as in the retail listings post.
2. Use Canadian prices and wording when you test. Our Canada prompts use Canadian dollars and Canadian phrasing. A US prompt is a different question.
3. Do not copy a US result. The overlap ranged from 15.8% to 88.9% across the 24 pairs, so a US check can mislead for your category.
4. Test each engine on its own. Excluding the alternative-to-a-brand question, the two engines' recommended brands overlapped by a mean of 45.9% per category in Canada, so one engine can name you while the other does not.
5. Look for your domain in the source lists. The Canada index category pages show the domains each category's answers drew on, for each engine. If you run your own checks, search for your .ca domain, any store subdomain and a parent company domain. The own-website post explains how strict matching misses some of them.

What this sample cannot tell you

• Developer APIs, not the consumer apps. What people see in the apps can differ through personalization, memory, location and interface choices.
• English only, once each. No French-language prompt was part of this run, and in the October 2026 edition each prompt got a single answer per engine.
• Answers, not behavior. We did not observe searches or purchases, so nothing here says where Canadians buy.
• No trend. The Canada categories are frozen, so there is no later run to compare with.

Frequently asked questions

Does a Canadian store in the list mean the assistant recommended that store?

No. Our measures count brands recommended in the answer text separately from domains in the source list, and stores are not counted as brands. A store domain in a list shows that the page was linked or retrieved.

Should a Canadian brand check its .com listings too?

It is worth a look. In this sample, amazon.ca was in the source lists of 93 Canadian answers and amazon.com in 14, both engines together, so the .ca listing appeared far more often but the .com one still appeared. These are counts of lists, not evidence of what shaped a recommendation.