We Checked 1,080 AI Answers for 18 Shopify Stores, Here Is What Got Cited

When shoppers ask an AI assistant what to buy without naming a store, small stores rarely appear: in our test, stores showed up in 9% of those answers, against 93% when the question named them. When a store's own site was cited, it was mostly a collection page or a product page. Most citations went elsewhere: Reddit, Amazon, YouTube and the big retailers. Here is the full picture, with the numbers.

What did we test, and how?

In August 2026 we took 18 independent Shopify stores and wrote 10 shopper questions for each: 54 category questions ("what are the best X to buy"), 89 use-case questions ("what should I get for Y"), and 37 brand questions that name the store or its signature product. We asked each question three times on two AI search engines, for 1,080 answers in total. None failed.

The two engines were Perplexity's Sonar model and OpenAI's search model, both called through their APIs with no chat history. These are the API versions, not the consumer apps, so what you see in the ChatGPT app can differ. We counted a store as appearing when its own site was in the answer's source list or the store was named in the text.

The limits are worth stating up front: 18 stores is a small sample, the whole test ran on one day, and we wrote the questions ourselves. Treat the numbers as a snapshot of how these two engines behaved in August 2026, not as fixed laws of answer engine optimization.

How often does a store appear when the shopper does not name it?

This is the number that decides whether AI assistants send you customers you do not already have. The split was stark.

Question type	Answers	Store appeared
Named the store	222	93% (207 of 222)
Generic, all	858	9.4% (81 of 858)
Generic, category	324	14% (45 of 324)
Generic, use-case	534	7% (36 of 534)

Every store appeared when the question named it, and the store's own site was cited in 71% of those answers: 84% on Perplexity, 59% on OpenAI's model. Being known worked. The engines could find, read and point at every one of these stores once a shopper said the name.

Generic questions were a different world. Eight of the 18 stores never appeared on any of them. Three stores appeared on one question, four on two, and one store each on three, four and six of their questions. Only 24 of the 143 generic questions produced any store appearance at all.

Which of a store's own pages get cited?

When an answer to a generic question did cite the store's own site, we logged which page it used. Across those 102 citations: collection pages 42%, product pages 27%, the homepage 17%, and blog posts or guides 13%, with one citation falling outside those groups.

The guide number needs careful reading. The 13 guide citations came from just 4 stores, 11 of them on Perplexity, and four of the five guides that got cited matched the question closely. That is too thin to conclude that guides are what gets cited, and it does not show guides fail either: the data cannot tell whether guides work less well or whether few of these stores had a guide that matched the question.

What the breakdown does say is that collection and product pages carried most of these stores' own citations. If an engine is going to point a shopper at your site, the page it reaches for first looks like a well-named collection.

Which other sites do the assistants lean on?

Most sources in generic answers were not store sites at all. A big retailer or marketplace appeared as a source in 40% of the 858 generic answers, and 2,274 different domains were cited across the test.

The two engines leaned on very different sites. Perplexity cited reddit.com in 83% of its 429 generic answers and amazon.com in 41%. OpenAI's model cited neither of those even once, and instead used youtube.com and walmart.com, each in 14% of its answers. Perplexity also cast a wider net, with a median of 17 source domains per answer against 7 for OpenAI's model.

One caution: this is an August 2026 snapshot. Source mixes like the 83% Reddit share move from month to month, so check them before you build anything on them. How the engines pick and rank sources is their own business, which also changes; our post on Google's AI Overviews covers what the vendors themselves publish.

Do ChatGPT's and Perplexity's models agree?

Rarely. Of the 24 generic questions where a store appeared, both engines showed the store on 6. Perplexity alone showed it on 11, and OpenAI's model alone on 7. Appearing on one engine said little about the other.

Within one engine on one day, though, results were steady: 353 of the 360 question and engine pairs gave the same result in all three runs, a 98.1% match. A check like this is stable enough within a day to be worth running, and different enough across engines to be worth running on both.

What should a store do with this?

Three moves follow directly from the numbers, and we wrote up the full playbook in how to get your store recommended by ChatGPT.

• Tend your collection and product pages first. They carried 69% of these stores' own citations (42% plus 27%), so their titles and intro copy deserve the attention blog posts usually get.
• Check both engines, not one. They agreed on 6 of 24 questions and leaned on different sites, Reddit and Amazon on one, YouTube and Walmart on the other. A store that only watches ChatGPT is blind to half the picture.
• Repeat the check monthly. Results were 98% stable within a day, but source mixes move over months, so one check is a baseline, not an answer.

Being known turned out to be different from being recommended: 93% when named, 9% when not. Closing that gap is slow work on pages, content and the places assistants read. If you want the same kind of test run on your own store, our free AI visibility check shows which shopper questions get your store named or cited by ChatGPT, Perplexity and Google AI Overviews.

What are the limits of this test?

Eighteen stores, ten questions each, one day in August 2026, and we wrote the questions. The engines were the API versions of Perplexity Sonar and OpenAI's search model, not the consumer apps, which can behave differently. The guide finding is directional at best, for the reasons above. And because the stores are real businesses, we report aggregates only and name no store.

FAQs

How often do AI assistants recommend small Shopify stores?

Rarely, when the shopper does not name them. In our August 2026 test of 18 stores, stores appeared in 9% of answers to generic shopper questions, and 8 of the 18 never appeared on any. When the question named the store, it appeared in 93% of answers.

What pages do AI assistants cite from a store's own site?

Collection and product pages first. Of the 102 times these engines cited a store's own site on a generic question, 42% were collection pages, 27% product pages, 17% the homepage, and 13% blog posts or guides.

Do ChatGPT and Perplexity cite the same sources?

No. In August 2026, Perplexity cited reddit.com in 83% and amazon.com in 41% of its generic answers, while OpenAI's search model cited neither and used youtube.com and walmart.com in 14% each. They agreed on only 6 of the 24 questions where a store appeared, and source mixes change month to month.