1. The Ordinary went from absent to present when the product fields were included
We asked: “Which fragrance-free facial moisturizers should I compare for a basic routine?” The collection retained 20 saved AI Overview answers to that one question. In four of those 20, the main saved text contained only an introduction that promised a list of products. Separate structured fields held product titles for La Roche-Posay, Vanicream, Aveeno, CeraVe and The Ordinary.
Looking only at the main saved text, CeraVe appeared in 15 of the 20 answers. Including the accompanying product fields raised that to 19 of 20. The Ordinary went from appearing in none of the 20 main texts to appearing in four of the 20 combined records.
Those are different brand stories. CeraVe was already visible in most of the stored text; the additional fields made its presence more complete. The Ordinary looked entirely absent from the prose measure, yet its product occupied a named slot in those four records.
The four records repeated one answer version. They are four saved occurrences, not four independently established choices. Across all 30 study questions, including structured product information recovered additional tracked names in 51 of the 681 selected answers. Those 51 answers were spread across 11 different questions. This recovered names; it did not establish 51 new product recommendations. [S08]
The useful distinction is between a brand missing from the answer and a brand missing from one captured text field. In this case, the record itself supplied evidence for the second explanation. It would be wrong to assume that explanation for every apparent absence.
2. Black Girl Sunscreen appeared in fewer answers, but more distinct versions
A second question asked which facial sunscreens to compare when white cast and a greasy finish were the main concerns. We retained 16 saved AI Overview answers to that one question. Counting the main text together with the accompanying structured content, Vacation and Colorescience each appeared in nine of those 16 answers. Black Girl Sunscreen appeared in five.
Some saved answers repeated the same wording. After grouping identical combined text, the 16 answers contained nine distinct versions; one version appeared eight times. Counting each version once changes the comparison. Vacation and Colorescience each appeared in two of the nine versions; Black Girl Sunscreen appeared in five.
Neither view erases the other. Vacation and Colorescience recurred more often in the collected records. Black Girl Sunscreen appeared across more distinct versions of the answer. The first describes saved-answer recurrence; the second shows how much that recurrence depends on repeated wording. [S09]
The same check also preserved a strong baseline: Supergoop and La Roche-Posay appeared in all 16 records and all nine versions. Repetition did not make every comparison unstable.
Collection also recorded 32 failed attempts for that sunscreen question. They are outside the 16 selected-answer denominator; a failed collection is not an answer that left a brand out.
For a marketer evaluating the competitive set, this matters more than declaring a corrected winner. A frequently repeated list and a brand appearing across several different lists are distinguishable findings. Different wording still does not prove independent retrieval, so the distinct-version view is a sensitivity check, not the one true ranking.
3. Platform links remained prominent after repeated text was grouped
For the final check, we looked across all 30 questions at links to social, video or forum platforms. Within each question, we calculated the fraction of selected AI Overview answers with at least one such link. Averaging those 30 fractions gave 83.8%, with every question contributing equally. This is an average rate of answers carrying a platform link, not the percentage of all citations that went to platforms.
We then grouped repeated main-text answers within each question and counted each group once, keeping any source families found in that group. The average rate was still 82.1%. Applying the same check to the study’s listed skincare-brand domains changed their rate from 37.8% to 32.7%. The broad platform pattern survived this repetition check. These source groups use main text; the sunscreen comparison above uses main text plus structured content. [S10]
The moisturizer record above makes the source mix tangible: its saved citation list included YouTube, Instagram and Facebook links alongside beauty publishers and The Ordinary’s website. That establishes which links accompanied the answer. It does not tell us which page supplied a product suggestion or which creator influenced it. [S11]
Some links could not be assigned to a source category: 31.5% of the counted website-host appearances across these answers were unclassified. That measure counts a host once per answer, so it is different from the question-average percentages above. A platform category is not a quality rating, and the study did not inspect every linked page. Nothing here proves that investing in a particular platform would improve a skincare brand’s AI visibility.
The three findings point in different directions. Complete fields recovered names. Repetition changed a comparison among sunscreen rivals. A broad source pattern persisted. That is the value of checking the underlying answers: sometimes the conclusion changes, and sometimes the evidence gives a clearer reason to keep it.
Methods and evidence
This is saved-evidence analysis of 681 selected AIO answers to 30 authored questions, collected October 9 and 12, 2026. Eighteen failed original runs were replaced by later runs of the same questions; their original partial answers were excluded. Capture remains incomplete. Product-name and source repetition checks use different representations, specified in the foundation methods. Results describe configured provider captures, not current search behavior, product efficacy or independent trials.
Sources: S07–S11: foundation claims, calculation tables and saved-record provenance.
Start with your own brand’s visibility. Try Gist’s free Quick Scan for a topic that matters to your business. It provides a snapshot of tested AI answers; the capture and repetition checks in this article are separate research work.
Read the full skincare study.
Skincare choices in AI answers includes the complete findings, methods and a 29-page PDF. Email capture opens access in this browser for one hour. No email is sent, and you won’t be subscribed to marketing.