AI names your brand when shoppers ask for a pick, and drops it when they ask what to look for

Most brand checks of AI answers start with the obvious question: "What is the best [product] for [need]?" That is a reasonable test. It is also the question where brands do best.

Two studies from the past month looked further back in the shopping journey, at the questions people ask before they ask for a pick. What should I look for in a dandruff shampoo? Which metals are safe for sensitive skin? Is this ingredient safe? On those questions, brands mostly disappear from the answer, and when they do appear, the answer rarely points to the brand's own website.

The Yarnit study: 1,100 answers, 56 brands

Between August and October 2026, the content company Yarnit put 550 shopper questions about 56 consumer brands to ChatGPT and Gemini. For each of the 1,100 answers, it recorded whether the brand was named and whether the answer linked to the brand's website.

The overall result is sobering. Across all answers, the engines named the brand being researched 28% of the time and linked to its site 11% of the time. On 62% of questions, neither engine mentioned it at all. Seven brands appeared in 65% or more of their answers. Six never appeared once.

The breakdown by question type is the finding we would put first.

Show the data

Question type	Value
Direct recommendation	37%
What to look for (criteria)	17%
Concerns	8%

Source: Yarnit, AI Shopping Visibility Report Fall 2026 (Oct 30, 2026). 550 questions about 56 consumer brands, asked of ChatGPT and Gemini, August to October 2026. Brand's own site cited in 2 of 108 concern answers.

Yarnit's examples of criteria questions are "what makes a sweater last" or "which metals suit sensitive skin." For packaged goods, think "what should I look for in a kids' sunscreen" or "how much protein should a snack bar have." Concern questions are things like whether an ingredient is safe or why a product causes irritation.

Yarnit's argument for why this matters is straightforward: the criteria a shopper picks up in the early questions decide which products make the shortlist later. And on concern questions, "where trust in a category forms," the engines cited the brand's own site in just 2 of 108 answers.

Two engines, two channels

The second finding is that ChatGPT and Gemini behave like different channels.

Measure	ChatGPT	Gemini
Times the brand was named	150	157
Share of mentions linking to the brand's site	70%	8%

Source: Yarnit. On more than half of the questions where a brand appeared, only one of the two engines named it.

So a brand can be well placed in one assistant and missing in the other for the same question. Measuring "AI visibility" as one number hides that.

The category view adds one more warning for CPG. Home and outdoor brands were the most visible, named in 39% of answers and cited in 18%. Beauty and personal care was the least visible of the larger segments: named in 16% and cited in 3%. Not one beauty brand made Yarnit's top group of "category definers," brands named in at least 30% of answers with their own site cited in at least 30% of those mentions. Yarnit's explanation is that the definers are brands an engine can describe in a phrase, like washable rugs or insulated drinkware.

The Euromonitor study: where the answers come from

Euromonitor International published a narrower but deeper test on October 7. It put 1,000 scalp-care prompts to ChatGPT and Gemini in the UK and the US, and looked at which brands appeared and which sources the answers cited.

The sources tell the story.

Source type	ChatGPT	Gemini
Retailer pages	Not given	58% of citations
Editorial and beauty media	32% of citations	Not given
Clinical sources	17% of citations	Not given
Brand-owned websites	3% or less	3% or less

Source: Euromonitor International. "Not given" marks shares the article did not give as a number.

Gemini also wrote more detailed product answers. It named the active ingredient and its concentration in 83% of answers, against 61% for ChatGPT. It explained how the ingredient works in 53% against 40%, gave pricing in 96% against 61%, and used review language in 45% against 8%. That pattern fits its heavy use of retailer pages, where prices, ingredients and reviews sit side by side.

Brand presence differed sharply by engine too. Nizoral appeared in 58% of Gemini's answers and 24% of ChatGPT's; Briogeo in 40% against 9%.

And it differed by question. Overall, Neutrogena appeared in 52% of answers, Nizoral in 41% and Head & Shoulders in 38%. But Head & Shoulders was in 66% of general dandruff answers and only 24% of answers about psoriasis or eczema, where Neutrogena reached 69%. The same brand can lead one need and fade on the next one over.

Country mattered as well. Philip Kingsley appeared in 48% of UK answers and 2% of US ones; CeraVe in 33% of US answers and 1% of UK ones. For a US brand checking its answers, testing in the US matters.

Reading the two studies together

The studies differ in size, method and categories, so we would not merge their numbers. But they point the same way on three things.

The shortlist is formed before the pick

Yarnit shows brands fading on early questions. Euromonitor shows answers changing by need within one category. Both suggest that a brand checking only "best X" questions is measuring the stage where it does best and missing the stages that set the criteria. We made a related point in our post on follow-up questions: the second and third question in a conversation are where products get filtered.

Your website is rarely the source

Brand sites were 3% or less of Euromonitor's citations, and they were cited in 2 of 108 concern answers in Yarnit's study. The engines read retailer pages, media and clinical sources first. That does not make your site irrelevant. ChatGPT linked to brand sites in 70% of its Yarnit mentions. It does mean the facts on retailer listings and in independent coverage carry most of the weight. Tip 30 covers retailer listings beyond Amazon.

Specific facts get repeated

Gemini's answers in the Euromonitor study named ingredients, concentrations and prices most of the time. Yarnit's definers are brands an engine can sum up in a phrase. The common thread is that assistants repeat what is stated plainly and specifically. That is consistent with Vercite's research and the other studies we covered in our post on where AI brand picks come from, from a different angle.

What we would do with this

1. Write down the early questions. For each product line, list five criteria questions and five concern questions shoppers actually ask. Customer service logs, retailer Q&A and reviews are good sources.
2. Check both engines separately. Run the list in ChatGPT and Gemini, in the US, and note whether you are named, what the answer says the criteria are, and which sources it cites. Some teams track this over time with tools such as SearchPanel from XENA Intelligence; a spreadsheet works for a first pass.
3. Answer the criteria on your own pages. If the answer says, for example, "look for zinc oxide" or "under 5 g of sugar," your page should state your number in the same terms. Tip 20 covers content for specific needs.
4. Fix the retailer listings first. If Gemini leans on retailer pages, those pages are your most quoted copy. Ingredients, amounts, sizes and claims should match your own site.
5. Earn the editorial and clinical citations. ChatGPT leaned on media and clinical sources in the Euromonitor study. Clear, citable information for reviewers and dermatologists or dietitians is part of this work, not a separate PR project.

Caveats

• Yarnit's 56 brands span fashion, home, accessories and beauty. Packaged food and household goods are not broken out.
• Euromonitor covers one category, scalp care, in two countries. It is detailed but narrow.
• Both studies used ChatGPT and Gemini only. Copilot, Perplexity, Claude and Google AI Mode were not tested.
• Assistant answers change with model updates and personalization. Both studies are snapshots from late summer and early fall 2026.

The short version: if you only test whether AI picks you, you are testing the last step. The steps before it, where shoppers learn what good looks like, are where most brands are absent.