What ChatGPT and Claude recommend in six DTC categories

On 23 September 2026 we asked consumer ChatGPT and Claude the same eight product questions in six DTC categories, once each, and recorded every brand they named. Across the 16 answers they named 66 distinct brands; 26 were named by both. Agreement swung from zero overlap on vitamin C serums to near-complete overlap on dog food. When ChatGPT cited a source, it was usually the brand's own website: 41 of the 53 distinct links it showed. This is a single-observer snapshot, not a ranking, and its limits are set out below.

We also asked both assistants nine questions a Shopify merchant might ask about AI search. Their answers are summarised in the last section, because they say something useful about where to spend effort.

The exact prompts and every brand mention are downloadable.

Monthly refresh. This page records one day's snapshot from the consumer apps. Its successor is the AI shopping index: its September 2026 edition asked ChatGPT and Perplexity three buyer questions in each of twelve DTC categories through an API, with CSV downloads. API answers are not the same as what a shopper sees in the apps, so compare the two as separate observations, not as a trend.

Findings at a glance

Question	Category	ChatGPT named	Claude named	Named by both	Distinct brands
Vitamin C serum for sensitive skin	Beauty	2	5	0	7
Small independent skincare brands	Beauty	4	7	1	10
Magnesium supplements	Supplements	5	5	4	6
Third-party tested protein powder	Supplements	7	6	5	8
Basic men's t-shirts	Apparel	8	6	4	10
Bed sheets	Home	7	6	4	9
Dog food	Pet	5	6	5	6
Specialty coffee	Food	7	6	3	10
Total		45	47	26	66

Counts are brands recommended in the answer text. A brand mentioned only to be ruled out is not counted.

Where the two assistants agreed, and where they didn't

Mature, evidence-led categories converged. For dog food both named Purina Pro Plan, Hill's Science Diet, Royal Canin, Eukanuba and Iams; Claude added The Farmer's Dog. For protein powder both named Thorne, Momentous, Klean Athlete, Optimum Nutrition and Garden of Life. For magnesium, four of six brands overlapped. In each of these categories both answers leaned on a third party: veterinary nutrition guidance for dog food, NSF Certified for Sport for protein, and product certification for magnesium.

Taste-led and niche categories diverged. For vitamin C serums for sensitive skin, ChatGPT recommended Vanicream and CeraVe, while Claude recommended Klairs, La Roche-Posay, The Ordinary, Mad Hippie and Beauty of Joseon. No brand appeared in both. For independent skincare the only shared brand was Stratia. For specialty coffee, three of ten brands overlapped: Onyx Coffee Lab, George Howell and Counter Culture.

They disagreed on facts, too. Both named Stratia, but ChatGPT placed it in Los Angeles and Claude in Seattle, and they gave its best-known moisturiser different names. Neither answer is evidence of which is right. The point for a brand is that assistants will state facts about you whether or not your own pages state them clearly.

Small DTC brands did get named. This was not only a list of household names. ChatGPT's coffee answer was built almost entirely from independent roasters: SEY, Onyx, Black & White, Heart, Sweet Bloom, George Howell and Counter Culture. Its independent-skincare answer named Stratia, Experiment, Dieux and Prequel. A small brand can be the answer when its own site explains clearly what it makes and why.

What ChatGPT cited

ChatGPT showed its sources as links. Across the eight answers there were 53 distinct links, and 41 went to a brand's own website: a product page, a collection or an about page. The others were editorial roundups (Allure, Sleep Foundation, a coffee guide), a certifier (NSF's product database) and public bodies (the NIH Office of Dietary Supplements, WSAVA, Tufts veterinary school, the FDA).

The split varied by category. In apparel all nine links were brand sites. For dog food and protein, none were: ChatGPT cited authorities and a certification database instead. That pattern suggests a practical rule. In categories where shoppers worry about safety, get certified and listed where the assistant looks. In categories where they choose on taste, make your own pages the clearest description of what you sell.

Claude's answers in the consumer interface showed sources differently. It mostly displayed merchant and source labels on product cards rather than inline links, so we don't report a comparable link count for Claude.

Product cards

Both assistants mixed product cards into some answers. ChatGPT showed them in five of eight answers and Claude in seven of eight. Three things stood out:

• Card text can be wrong. One ChatGPT card described Experiment's Super Saturated, a 30% glycerin hydrating serum according to the same answer's text, as a "high-protein supplement for muscle recovery". ChatGPT's documentation says card titles and labels are generated by ChatGPT and "not guarantees or verified statements" (ChatGPT Help Center, Shopping with ChatGPT Search, read 23 September 2026).
• Claude's cards often showed resellers. Marketplaces and third-party sellers appeared as sources far more often than the brands' own stores.
• Location overrode the prompt. Every prompt said "I'm shopping in the US", yet ChatGPT's cards priced products in rupees, the observer's local currency. Anyone testing their own visibility should test from the market they sell to.

For Shopify stores, cards in ChatGPT draw on Shopify Catalog. How that works, and the one setting to check, is in Shopify agentic storefronts and ChatGPT Shopping.

What the assistants told merchants

We then asked each assistant nine questions a Shopify merchant might ask, such as "How do I get my Shopify store to rank in ChatGPT answers?" and "Does llms.txt actually help SEO or AI search visibility?". Their answers were consistent on four points:

1. Check Shopify's agentic storefronts first. Asked how to rank in ChatGPT, both made catalog access the first step. Claude opened with "Make sure your store is actually in ChatGPT's catalog"; ChatGPT with "Make sure Shopify → ChatGPT product discovery is enabled". Both pointed to Sales channels > Agentic in the Shopify admin.
2. Product data is the lever. Both said complete titles, attributes, variants, prices and policies matter more than any trick.
3. llms.txt is not a visibility lever. ChatGPT called its value "very low / speculative"; Claude said "on current evidence, no". Our llms.txt guide has the underlying studies.
4. Allow the search crawlers, decide training separately. Both separated OAI-SearchBot and Claude-SearchBot, which affect search answers, from GPTBot and ClaudeBot, which collect training data.

Asked for the best AI SEO app for a Shopify store, ChatGPT recommended StoreSEO and SearchPie. Claude said there is no single best app and cautioned that most "best app" lists are written by app makers. Neither named Citelift. We report that because it is what we observed. It is also the gap Citelift exists to close for its own customers: being absent from the answer until pages worth citing exist.

Method and limits

• Collection. 23 September 2026, by one observer (Siddesh Patil) in a desktop browser. ChatGPT was used in a temporary chat on the Instant setting. Claude was used in an incognito chat on its default model at the time. Each question was asked once per interface.
• Location. The observer was in India. Prompts stated "I'm shopping in the US", but results still showed local signals, including rupee prices on ChatGPT's cards.
• Context carry-over. Some Claude answers referred to "your earlier question" despite incognito mode, which suggests context carried between chats in the same session. Merchant-side answers may be more Shopify-specific than a cold start would give.
• Counting. A brand counts when the answer recommends it. Link counts are distinct URLs visible in ChatGPT's answer, with query strings removed.
• No repeats. Answers vary between runs. Our repeatability pilot found no identical source sets across 48 repeated answers. Read every number here as one observation.
• No commercial relationship. None of the brands named is a Citelift customer, and Citelift sells none of their products.

Data and reuse

Two files are available:

• prompts.csv: the eight product questions, verbatim.
• brand-mentions.csv: one row per brand per interface, with the order named and whether the other interface also named it.

The brands-per-question chart is drawn from brand-mentions.csv; the link chart uses the link counts reported above. For the same kind of data refreshed every month, see the AI shopping index.

Raw answer text stays private because it contains third-party content. If you run the same questions and get different results, that is expected, and we would like to hear about it at support@citelift.app. To run a smaller version for your own category, the method is in how to measure AI visibility for your store.