What ChatGPT and Gemini read before recommending a beauty brand
We asked ChatGPT and Gemini 42 beauty shopping questions, 420 answers in total. Which sources they used, how often they searched the web, and what it means for independent brands.
Short answer: ChatGPT searched the web before almost every beauty recommendation and leaned on Allure, the American Academy of Dermatology, Healthline and the big retailers. Gemini mostly answered from what it had already learned, and named retailers far more often. Independent brands received about half of all brand mentions, so the answers are not reserved for the giants.
How we measured
- 42 shopper questions in beauty and personal care, written as US shoppers ask them: by concern, ingredient, product type, budget and 'which smaller brands are worth it'.
- Two engines: ChatGPT (OpenAI API with web search, US location) and Gemini (Google API with Google Search grounding, told the shopper is in the US).
- Each question asked 5 times on each engine: 420 answers in October 2026.
- Brands extracted from every answer, counted as 'named in X of N answers', with 95% intervals.
Finding 1: ChatGPT searches, Gemini remembers
In our main scan of 280 answers, ChatGPT searched the web in 122 of 140 answers (87%). Gemini searched in 7 of 140 (5%) and otherwise answered from its training data. On narrower, more specific questions Gemini searched more often (20 of 70 answers), but still far less than ChatGPT.
This matters because the two engines reward different things. To be named by ChatGPT, a brand needs to appear in the pages it finds today. To be named by Gemini, a brand needs to have been widely written about before the model was trained.
Finding 2: the sources ChatGPT trusted
| Source | Citations in ChatGPT answers |
|---|---|
| allure.com | 201 |
| aad.org (American Academy of Dermatology) | 83 |
| healthline.com | 37 |
| sephora.com | 33 |
| ulta.com | 32 |
| ada.org (American Dental Association) | 24 |
| goodhousekeeping.com | 22 |
| consumerreports.org | 16 |
Two patterns stand out. Editorial lists (Allure, Good Housekeeping) shape which brands are considered at all. For anything touching skin conditions, expert bodies such as the AAD set the criteria, and ChatGPT then recommends brands that match those criteria.
Finding 3: Gemini names retailers
Gemini's answers named where to buy far more often than ChatGPT's: Target (99 mentions), Amazon (89), Ulta (87), Sephora (71), Walmart (53) and CVS (42). For brands, retail distribution is part of AI visibility, not only a sales channel.
Finding 4: independent brands get about half the mentions
Across all answers, mentions split roughly evenly between large mass-market brands and independent ones (ChatGPT 298 vs 307; Gemini 411 vs 475, using a rough classification). Head brands were very stable: CeraVe was named in 10 of 10 answers for very dry, itchy skin. But the long tail changed from one run to the next, which is where smaller brands can win.
Finding 5: specific questions are where smaller brands win or lose
On specific ingredient and format questions, small brands appeared, or were conspicuously absent. For 'best copper peptide serum', two brands were named in every answer, while a smaller brand that sells 2% and 3% copper peptide serums did not come up. Its product names use the ingredient's technical name, 'GHK-Cu', rather than the phrase shoppers use.
What this means for an independent brand
- Use the customer's words on product pages and titles, not only technical or brand terms.
- Get into the sources AI reads for your category: the editorial lists, expert guidance and directories it cites.
- Make facts checkable: certifications with links, ingredient lists, concentrations, test results, as text.
- Measure both engines. Being strong in Gemini's memory doesn't mean ChatGPT can find you today.
One AI answer is an anecdote. Ten answers per engine, counted the same way, is a measurement.