On this page
- What to Buy Is Answered From Store Listings and Publishers' Picks
- Questions About How Products Work Go to Regulators, Testers and the Makers' Guides
- ChatGPT Cites Makers' Sites as Often Either Way, From Different Pages
- The Sites Behind Buying Answers Reach Fewer Than Half the How-It-Works Answers
- Shoppers Check What the Assistant Told Them, Often on the Maker's Own Site
- The Listings Team Owns One Half, and Support Owns the Other
- Frequently asked questions
Two things to hold while reading. A citation shows which pages an answer pointed to, not that it named your product. And every share below counts the answers that cited at least one website, per engine and per kind of question.
Your product listing can miss shoppers who ask how something works before choosing a brand. In Adobe's July 2026 survey of US consumers, 42% had used an AI assistant for online shopping; 53% of that use was for inspiration and discovery, and 1% came after purchase.1 In Adobe's 2025 survey, research was the most common shopping task people gave the assistant, ahead of asking for product recommendations.2
That early asking comes in two kinds. “Best cordless vacuum for a small apartment” is a buying question: it asks which product to get. “How do cordless vacuums compare to corded ones for everyday cleaning” is a how-it-works question: how a product works, how two kinds differ, how long one lasts, whether a feature is worth paying for. It names no brand at all.
Ask whether the assistants recommend your company and the natural test is the first kind. On October 19 we put the same shopper questions to ChatGPT and Google AI Mode, Google's chat-style search answers, in the US: buying questions across six product categories, from electronics to baby gear, and the how-it-works questions people ask around them.
Buying answers are built from stores' product pages and publishers' best-of lists. How-it-works answers are built from government and standards pages, Consumer Reports, and the makers' own explanation pages: guides, comparisons, FAQs and support articles. Your support desk has a saved reply for how long the filter lasts, what the warranty covers and whether the jacket survives a wash. It's been writing the kind of page these answers cite for years without anyone calling it that.
What to Buy Is Answered From Store Listings and Publishers' Picks
Publishers appeared in 85.47% of ChatGPT's buying answers that cited a source, and stores and marketplaces in 70.39%. Walmart alone was in 42.46%, Target in 20.67%, Good Housekeeping in 19.55% and Consumer Reports in 18.44%. The store links are product pages: the listing itself, not the store's home page.
On ChatGPT, buying answers cite stores and lists; how-it-works answers cite government pages and makers' guides
% of cited answers
View the data
| Label | Buying questions | How-it-works questions |
|---|---|---|
| Publishers | 85.47% | 70.83% |
| Stores and marketplaces | 70.39% | 25.00% |
| Government, standards and research | 19.27% | 49.31% |
| Makers' explanation pages | 1.96% | 29.86% |
| Makers' product and home pages | 36.03% | 8.33% |
| Forums, video and social | 9.78% | 0.69% |
OpenAI describes the mechanism in its own help pages. Product results are “selected independently by ChatGPT and are not ads,” built from “structured metadata from first-party and third-party providers” (the product details stores and makers supply in a set format), with review summaries “based on reviews from public websites.”
When several stores sell the item, merchants are ranked partly on “whether they are the maker or primary seller.”3 By that description, a maker's own store competes with Walmart's for the same link.
Google AI Mode builds the buying answer differently. 51.67% of its buying answers showed Google's own product cards, drawn from Google's database of product listings. Beside the cards, publishers appeared in 77.83% of the buying answers that cited a website, led by The New York Times's Wirecutter at 26.70%. Reddit followed at 18.10% and YouTube at 15.84%. Makers' own sites were in 4.52%.
None of this is news to whoever runs the units you send reviewers and your retailer listings. Getting onto the best-of lists and keeping the product data straight where ChatGPT reads it are the work on the buying side, and so is the product data file you send to Google and the stores, which feeds Google's cards.
Questions About How Products Work Go to Regulators, Testers and the Makers' Guides
Government, standards and research pages appeared in 49.31% of ChatGPT's how-it-works answers: Energy Star, the EPA, the FTC's consumer pages, the FDA, the Consumer Product Safety Commission, the NIH's research library. Consumer Reports alone was in 29.17%. Stores fell to 25.00%, and much of what stayed was the stores' own advice: REI's Expert Advice pages, Home Depot's and Lowe's buying guides.
The makers' pages that got cited were explanations, and several sat in support centers rather than on product pages: asked about down against synthetic insulation in jackets, ChatGPT cited Patagonia's own guide to the two fills.
On Google AI Mode, how-it-works answers lean on makers' explanation pages and forums
% of cited answers
View the data
| Label | Buying questions | How-it-works questions |
|---|---|---|
| Publishers | 77.83% | 38.78% |
| Stores and marketplaces | 7.24% | 26.53% |
| Government, standards and research | 1.36% | 30.61% |
| Makers' explanation pages | 2.71% | 51.02% |
| Makers' product and home pages | 1.81% | 10.20% |
| Forums, video and social | 35.75% | 48.98% |
On Google AI Mode the makers lead. A maker's explanation page appeared in 51.02% of the 49 how-it-works answers that cited a website.
“Is it worth buying linen sheets over cotton” was answered from bedding makers' own linen-versus-cotton pages, and whether air purifiers help with allergies from purifier makers' pages, next to the American College of Allergy, Asthma and Immunology's.
Reddit (32.65%) and YouTube (24.49%) sat beside them, and government pages appeared in 30.61%.
Google AI Mode also left most of its how-it-works answers without a source. It cited a website in 34.03% of its how-it-works answers, against 61.39% of its buying answers; the rest answered from nothing it linked.
The how-it-works shelf on Google is smaller, and the makers hold more of it than any other kind of site.
ChatGPT Cites Makers' Sites as Often Either Way, From Different Pages
On ChatGPT, a maker's own site was cited about as often either way: in 37.43% of its buying answers and 36.11% of its how-it-works answers.
What changed was the page. In buying answers, 36.03% cited a maker's product, category or home page, and almost none an explanation page. In how-it-works answers, 29.86% cited an explanation page and 8.33% a product page.
Ask which clothing brands are most recommended and you get Uniqlo's and Levi's home pages; ask how to tell whether a sofa will hold up and you get a sofa maker's list of questions to ask before buying.
How often ChatGPT linked the maker on a buying question depended on what you sell:
ChatGPT links makers' sites most on electronics questions and least on baby products
% of cited buying answers
View the data
| Label | Value | Note |
|---|---|---|
| Electronics and gadgets | 48.19% | |
| Home and kitchen | 42.86% | |
| Apparel and footwear | 38.33% | |
| Fitness and outdoor gear | 35.56% | |
| Beauty and skincare | 29.63% | |
| Baby products | 9.38% | Baby buying answers went to stores (93.75%) and to pediatric and medical sources (40.62%). |
Keep buying questions and how-it-works questions separate: they test different pages of your site.
The Sites Behind Buying Answers Reach Fewer Than Half the How-It-Works Answers
Take the short list of sites that together appear in four out of five buying answers. On ChatGPT they reached 40.28% of the how-it-works answers; on Google AI Mode, 48.98%.
The sites behind buying answers reach under half the how-it-works answers on both engines
% of cited answers
View the data
| Label | ChatGPT | Google AI Mode |
|---|---|---|
| Buying-answer sites, on how-it-works answers | 40.28% | 48.98% |
| How-it-works sites, on buying answers | 71.79% | 57.92% |
The overlap runs one way. The how-it-works list reached 71.79% of ChatGPT's buying answers and 57.92% of Google AI Mode's. A few broad sites sit on both lists: Consumer Reports, Good Housekeeping, TechRadar and Walmart on ChatGPT; Wirecutter, Reddit and YouTube on Google.
A source plan built from buying questions misses more than half the how-it-works answers and the guides they cite.
One month's screenshot is not a measurement. In an independent September 2026 audit by university researchers, ChatGPT stated a first-person product preference in 79% of the answers that recommended a product, and the products it recommended often changed when the same request was repeated.4
Shoppers Check What the Assistant Told Them, Often on the Maker's Own Site
The explanation page has a second reader: the shopper who clicks through to check. 66% of consumers in Adobe's July survey said they visit a brand's website directly to verify an AI recommendation, and 89% said they click the links or sources an assistant provides.1
In Gartner's survey of shoppers who had used AI for a recent purchase, 54% had to double-check the accuracy of all the information it gave them, and 62% said it ended up a waste of their time.5
The checking doesn't hand the decision to the assistant. Gartner found willingness to let AI make the purchase decision topped out at 11%, and only in low-stakes categories such as personal care and household supplies.5 The shopper decides, and decides partly on what your own pages say.
AI-referred visits to US retail sites converted at a rate 60% higher than other traffic in July 2026, and AI-driven retail traffic grew 62% year over year, in Adobe's data.1 This covers all AI-referred visits, not explanation pages alone.
The Listings Team Owns One Half, and Support Owns the Other
Each half runs through work someone in your company already does:
| Question | What the answers read | Who owns it |
|---|---|---|
| What to buy | Store product pages, ranked lists, Google's product cards | Whoever runs listings, review units and the feed |
| How it works | Government pages, Consumer Reports, makers' guides | Support, with the product team |
Vlad ShvetsPartner @ QveryThe how-it-works answer is where your own explanation can stand beside the government page and the tester. Check the saved reply against the product spec sheet, then have support and product approve it for publication, because an email answer still needs that check before it becomes a public claim. If your support team already explains it well by email, the page is half written.
So the buying half belongs to whoever runs your listings, the units you send reviewers and the product data file; the how-it-works half belongs to the support desk, with the product team checking the facts. Each needs the other's pages measured on the same questions every month.
Frequently asked questions
Which websites do AI engines cite when shoppers ask for product recommendations?
In our research, publishers appeared in 85.47% of ChatGPT's buying answers that cited a source, and stores in 70.39%. Google AI Mode cited publishers in 77.83% of buying answers citing a website; its product cards appeared in 51.67% of all buying answers.
Do AI engines cite a brand's own website?
Yes, and the page depends on the question. ChatGPT cited a maker's own site in 37.43% of buying answers and 36.11% of how-it-works answers: product and home pages for the first, guides, FAQs and support articles for the second. On Google AI Mode, makers' explanation pages appeared in 51.02% of the 49 how-it-works answers that cited a website and 2.71% of buying answers.
What do AI answers cite when shoppers ask how a product works?
Government and standards pages, Consumer Reports and makers' explanation pages. On ChatGPT, government, standards and research pages appeared in 49.31% of how-it-works answers and Consumer Reports in 29.17%. On Google AI Mode, makers' guides led, with Reddit (32.65%) and YouTube (24.49%) beside them.
Do ChatGPT and Google AI Mode cite the same sites for shopping questions?
Not for the same kind of question. ChatGPT's how-it-works answers lean on government pages and Consumer Reports and almost never cite a forum. Google AI Mode's lean on makers' pages, Reddit and YouTube, and it cited no website at all in most of its how-it-works answers: it cited one in 34.03%.
Does a citation mean the assistant recommended our product?
No. A cited page is what an answer was built from. An answer can cite your guide and recommend someone else's product, or name your product while citing a store's listing, so the brands an answer names are counted separately from the pages it cites.
Is traffic from AI answers worth anything to an online store?
On Adobe's numbers, yes. AI-referred visits to US retail sites converted at a rate 60% higher than other traffic in July 2026, and AI-driven retail traffic grew 62% year over year.1 66% of consumers say they visit a brand's website to verify an AI recommendation.1
Who in our company should own how AI answers describe our products?
Two people. Whoever runs retailer listings, the units you send reviewers and the product data file owns the buying half. Whoever answers customers' product questions, with the product team, owns the how-it-works half, because the pages those answers cite are explanations that support already writes by email.
How often should we check what AI answers cite?
On a fixed schedule, with the same questions each time, and more than once per question. A September 2026 audit found the products ChatGPT recommended often changed when the same request was repeated.4
Do we need to send our product feed to ChatGPT?
It helps the buying half. OpenAI says ChatGPT builds product results from structured metadata from first-party and third-party providers, and that merchants on Shopify are already integrated through Shopify Catalog.3 The how-it-works half runs on your explanation pages instead.
Which stores and publishers does ChatGPT cite for buying questions?
Walmart most: it appeared in 42.46% of ChatGPT's buying answers that cited a source, Target in 20.67%, Good Housekeeping in 19.55% and Consumer Reports in 18.44%. The store links went to product pages, not to the stores' home pages.
Which sites does Google AI Mode cite for buying questions?
Publishers first: The New York Times's Wirecutter appeared in 26.70% of its buying answers that cited a website, Reddit in 18.10% and YouTube in 15.84%, while makers' own sites were in 4.52%.
Which product categories get the maker's own site cited most?
On ChatGPT's buying answers, electronics: a maker's own site was cited in 48.19% of electronics answers and 9.38% of baby-product answers, where stores and medical sources filled the answers instead.
Sources
- Adobe: AI Traffic Trends Report (2026)
- Adobe: Adobe Analytics: Traffic to U.S. retail websites from Generative AI sources jumps 1,200 percent (2025)
- OpenAI: Shopping with ChatGPT Search (2026)
- arXiv: If I Had to Buy Just ONE: Auditing AI-Generated Product Recommendations (2026)
- Gartner: Gartner Survey Finds Consumers Want AI Shopping Help, But Not AI Purchase Decisions (2026)