A handful of brands per category dominate the recommendations generative AI chat platforms give, according to a new report from ecommerce platform CommerceV3, which also sells AI visibility services.
The report found that AI platforms filled 75% of their recommendation slots with roughly the same five brands, CommerceV3 CEO Rahul Amarendra said in an email. In fact, 78% of the 25,000 ecommerce brands it analyzed never appeared in AI recommendations in their own category.
The report found that ChatGPT, Gemini and Perplexity frequently cite category leaders, marketplaces and editorially validated brands, while others go unmentioned.
CommerceV3 analyzed more than 1 million AI shopping prompts across ChatGPT, Gemini and Perplexity between March and July 2026 in the U.S. The analysis ran each shopping prompt five times per engine and measured which brands the model cited or recommended in its response. CommerceV3 counted a brand as visible only if it appeared in at least two of the five runs. Example prompts included “best running shoes for flat feet” and “compare Brand A vs. Brand B.”
The analysis included seven ecommerce categories, about 200 subcategories and roughly 25,000 ecommerce brands, which CommerceV3 compiled from public marketplace listings, retail directories and category-level web commerce indexes.
Each platform sources results slightly differently. On Perplexity, which relies on live web citations, brands were absent from 74% of category prompts (a 74% “brand absence rate”), compared with an 81% brand absence rate on Gemini, which draws on Google’s index and Shopping Graph. ChatGPT, which leans on its training corpus and browsing, had a 76% brand absence rate.
Brands were most likely to be absent from problem-solving prompts (84%) and broad recommendation prompts (81%). Categories with strong editorial review ecosystems, such as consumer electronics and beauty, had a lower brand absence rate.
In specialty and long-tail categories, such as hobby and grocery, the brand absence rate was much higher.
Not being cited hurts companies more than a low Google ranking would, according to the report. On Google, a brand that ranks in the bottom half of the first page, or on Page 2 or 3, still receives some residual traffic.
“An assistant recommends three to five options; every other brand receives nothing — no impression, no click, no signal it was even considered,” the report said.
This is important for brands as more consumers use these platforms to shop. In July 2026, visits to U.S. retail sites from AI sources increased 62% year over year, according to data from Adobe.
Where Generative AI Platforms Find Information
The majority of AI’s recommendations point to third-party sources, including editorial reviews, forums and marketplaces.
AI cites the following sources:
- Editorial and review sites — 34% share of citations
- Reddit and forums — 22%
- Marketplaces — 18%
- Brands’ own sites — 14%
- Video content — 7%
- Other — 5%
Because generative AI sources information from brands’ own websites only 14% of the time, organizations need to make sure third-party sites are talking about them.
Brands that AI cites share similar structural traits. Their content includes “extractable answer content” — concise, question-shaped pages that generative AI can lift answers from directly — and schema markup that helps AI read the content.
For example, an AI chatbot can source information on a product page such as “This leave-in conditioner is safe for color-treated hair, it’s sulfate-free and pH-balanced” instead of “a revolutionary formula that transforms your hair,” Amarendra said.
Brands AI regularly cites clearly state who they are and what they do, so the chatbot doesn’t have to rely on context clues. These brands also keep facts about themselves consistent across the internet.
This means companies that invest time and talent can improve their AI visibility without relying on a bigger ad budget, according to the report.
“The brands winning those slots aren’t the biggest spenders,” Amarendra said in an email. “We looked for that and didn’t find it. What they have in common is content the AI can quote directly, clean product data and schema, and plenty of mentions on the review sites, forums and marketplaces the AI leans on.”
