We asked AI assistants about five real online stores. Here is what decided the answers

By 7BE ResearchPublished 25 October 2026

Across five independent online stores, ChatGPT, Perplexity and Google AI Mode named the store in 11 of 84 answers to questions its shoppers might ask. Well-known brands and big retailers filled most answers, backed by magazines, review sites and retailer pages. The one store that did well makes acoustic panels, a small category with few rivals, and the wording of the question changed the answers more than we expected.

We ran these scans on 24 October 2026. It is one run on five stores, so read the numbers as a close look at real answers from one day. The method is at the end. We do not name the stores.

How often did AI assistants name each store?

Answers are the answers we received to each store's 8 shopper questions. Named means the answer mentions the store or links to its site. Readiness is the store's score out of 100 on the 8 readiness checks. The data provider refused some calls, so the number of answers differs per store.

Four of the five stores were missing from at least four in five of their answers. The skincare brand's one mention was a row in a ChatGPT table of other professional brands worth comparing, below the brands it described in full.

Who did the assistants recommend instead?

Counting the answers that mention each brand or retailer:

• Fragrance (22 answers): Phlur in 9, Henry Rose in 6, Sol de Janeiro in 5, and the retailers Sephora in 8 and Ulta in 7.
• Professional skincare (23): Dermalogica and La Roche-Posay in 6 each; PCA Skin, SkinCeuticals, Circadia, IMAGE Skincare and BABOR in 5 each. Consumer questions, such as a serum for acne-prone skin, brought drugstore names like La Roche-Posay and CeraVe. Questions for estheticians brought the professional lines.
• Coffee machines (16): Breville and De'Longhi in 10 each, Philips in 7. Among retailers, Myer appeared in 4, and Appliances Online, The Good Guys and Amazon in 3 each.
• Modern furniture (13): EQ3 in 7, Structube in 6, IKEA, Wayfair and Crate & Barrel in 5 each, Article in 4, West Elm in 3. Asked for modern furniture stores in the store's own city, all three assistants listed other shops and left it out.
• Acoustic panels (10): Naturewall in 9, three times in each assistant.

EQ3, Breville, De'Longhi, Dermalogica and Naturewall were each named by all three assistants. ChatGPT pointed to big retailers more often: in the fragrance answers it mentioned Sephora and Ulta 5 times each, while Perplexity mentioned Sephora once and Ulta never.

Which sites do AI shopping answers cite?

Leaving out Google's own product listings and ChatGPT's search links, the 84 answers cited 611 links on 309 different sites. The ones that came up most, by number of answers:

• Magazines and review sites: Allure (16), Vogue (9), Elle (5), Marie Claire (5), Good Housekeeping (4), Byrdie (4), the Australian consumer group CHOICE (4) and TechRadar (3).
• Retailers and marketplaces: Ulta (12), Sephora (9), The Good Guys (4), Appliances Online (4), Wayfair Canada (4), Best Buy Canada (3), Walmart (3) and Target (3). Amazon pages appeared in 6.
• Brand sites: BABOR (6), EQ3 (5), Structube (5), Naturewall (5), Phlur (4) and IKEA (4).
• Health sources: the American Academy of Dermatology (4), all in skincare answers.
• Forums: fewer than we expected. Reddit appeared in 3 answers, all from Perplexity.

The scanned stores' own sites were cited in 1 of the 84 answers.

The assistants also cite differently. ChatGPT and Perplexity listed about 10 sources per answer. In Google AI Mode, 45 of the 85 links pointed to Google's own product listings, which draw on Google's Shopping Graph and the product data merchants send to Google Merchant Center. For Google AI Mode, your product feed may count as much as any article about you.

Why did the acoustic-panel maker do better?

The UK acoustic-panel maker was named in 6 of its 10 answers, by all three assistants. In two ChatGPT answers its panels were the first pick. We see three likely reasons, and one run cannot prove any of them:

1. A small category. The answers drew on a short list of specialist panel makers and DIY chains. No global brand appeared in them the way Breville and De'Longhi appeared in the coffee answers.
2. Many pages mention it. The answers linked its panels to several retailers, including DIY chains, so the range appears on many of the pages the assistants read. One Google AI Mode answer described it as the first of its kind in the UK, and its product listing showed a high rating from hundreds of reviews.
3. A readable store. It passed 7 of the 8 readiness checks and scored 90.

There is a catch. None of those 10 answers cited the maker's own site. The links went to retailers that stock its panels and to Google's product listings, so a shopper who clicked would have landed on a retailer.

Does a higher readiness score mean more mentions?

Not in this sample. The coffee-machine retailer scored 95, the highest of the five, and was named in none of its 16 answers. The furniture store scored 55, the lowest, after failing the reviews, descriptions and policies checks, and was also named in none.

The checks decide whether an assistant can read and trust a store; they do not put it on the pages the answers are built from. The coffee retailer sells other companies' machines, so questions about the best machine were answered with the makers' names, and the question about where to buy with other retailers, mostly large national ones.

The checks still remove reasons to leave a store out. Asked about the furniture store by name, ChatGPT and Perplexity both said its independent reviews were limited, and the store had also failed our reviews check.

The assistants knew every store by name

In an earlier run the same day, the assistants answered both brand questions for all five stores: whether it is a good brand, and whether it is legit. All 30 answers recognised the store, and all 15 answers to the legit question called it legitimate, some with a caveat about limited independent reviews. 18 of the 30 cited the store's own site.

Compare that with one citation of a store's own site in 84 answers to shopper questions. The assistants know these stores exist. When the shopper named only a need, they reached for bigger brands and retailers, the ones the cited pages wrote about.

Why does question wording matter so much?

Our first run that day built each store's 8 shopper questions from its catalog tags and product types, in the form “What are the best [tag] to buy right now?”. Tags are written for a store's back office, and it showed: “best 90mls”, “best retails”, “best packs”, “best controls”.

The assistants asked what we meant, or guessed. “Best retails” brought retail stocks in all three. “Best controls” brought gaming controllers. Perplexity read “best 240mls” as computer CPU coolers, and one answer to “best buffets” covered Las Vegas buffets. By our reading, 57 of the 97 answers were about a different kind of product from the store's main range, or asked us to clarify. The stores were named in 2 of the 97, both times in answer to a question that contained the store's own name.

With questions written the way shoppers ask, such as “best vanilla perfume under $150” or “best acoustic panels for a home office”, all 84 answers were about the right kind of product, and most named real competitors. When you test, write the question as a customer would say it, with the need, the budget and the country, and leave your store's name out of category questions.

What this means for your store

1. Measure with shopper questions. Ask in all three assistants, and more than once, because answers change between runs.
2. Read the source lists. The sites cited for your category are where a mention counts: the guides, magazines and retailers the assistants rely on.
3. Look at who takes your spot. The brands named instead of you, and the reasons given for them, show what your pages and listings are missing.
4. Pass the readiness checks anyway. Start with reviews: asked about a store by name, the assistants said so when independent reviews were thin.

The free 7BE scan does the first three for your store, with 10 shopper questions in ChatGPT, Perplexity and Google AI Mode, and runs the 8 checks. It shows the brands named instead of you and ranks the fixes. For the bigger picture, the AI Shelf Index asks 50 shopper questions in each of 10 categories.

How we ran the test

• Stores: five independent stores, three on Shopify and two on WooCommerce, in the US, Australia, Canada and the UK.
• Questions: 8 hand-written shopper questions per store, plus 2 brand questions. Each was asked once in ChatGPT, Perplexity and Google AI Mode through a data provider's API on 24 October 2026: 150 calls.
• What counts: the provider refused 65 calls and one came back empty, which left 84 answers to shopper questions. It refused every brand call, so the brand results come from the earlier run that day, which asked the same two brand questions.
• Counting: we counted brand mentions ourselves, once per answer. Source counts leave out Google's product-listing links and ChatGPT's search links.