Google's AI Mode linked only to Google in 46.5% of US product answers. Findrix's data, read for CPG brands

If a shopper asks Google's AI Mode for a reef-safe sunscreen, where does the answer send them? According to a new study, about half the time it sends them nowhere but Google.

Findrix, a company that measures AI visibility, put 18,400 buyer questions to ChatGPT, Gemini and Google AI Mode across 35 US markets between October 14 and 17, 2026, and collected 55,198 answers. Search Engine Land reported the headline result on October 29: AI Mode "linked only to Google's own pages in 46.5% of U.S. product answers," and 57% of all links in its product answers led to Google Shopping product pages.

Findrix also publishes the underlying numbers by market and by engine in its US Radar. Nine of its 35 markets are consumer goods categories where packaged goods brands compete: haircare, makeup, sun care, facial skincare, oral care, sports nutrition, pet food, and several apparel markets. Below we pull out what matters for those brands.

What Findrix measured, and what it did not

The method is simple and worth knowing before reading the numbers. Findrix wrote buyer questions for each market, about 526 per market, in the forms people use: comparisons, "best for" questions, problems and price questions. Each question was asked once to each engine. Findrix then counted which brands were named and which domains were linked.

That makes it a measure of what an answer shows, not of what shoppers do next. Search Engine Land is explicit: "The findings measure links shown in AI answers, not actual clicks or purchases." Each question was also asked only once per engine, and AI answers vary from run to run, so treat the category numbers as a snapshot of four days in October. Findrix publishes a 95% range for each brand's mention rate for the same reason.

AI Mode keeps product answers on Google

The category split is where it gets specific. AI Mode kept answers on Google most often for fashion and beauty products, and much less for sports nutrition.

Show the data

Market	Value
Swimwear	60.4%
Women's everyday fashion	57%
Sun care	54.1%
Activewear	53.8%
All US product answers	46.5%
Sports nutrition	7.6%

Source: Findrix study via Search Engine Land, October 29, 2026. Links shown in answers, not clicks. Each question asked once.

Findrix's co-founder Yosif Merman attributed the difference to Google's product database, which holds prices, ratings and inventory. With that data on hand, AI Mode can build an answer with fewer links to outside websites. Sports nutrition is the odd one out, and the study does not explain why. One plausible reading is that supplement questions lean on ingredients, dosing and safety, which pushes the engine toward health and editorial sources rather than shopping listings, but that is our guess, not a finding.

For a sun care or skincare brand, the practical meaning is plain. When AI Mode answers a product question in your category, the most likely destination is a Google Shopping product page, and the facts on that page come from the product data merchants send to Google. Your website may never be linked. Your feed will almost certainly be read.

ChatGPT links out, and leans on retailers and Reddit

The other two engines behave very differently. ChatGPT included about 25 links per answer, against about three for AI Mode. In product answers, it linked to Walmart in 45.1% of cases and to Amazon in 6.1%. It cited Reddit in 35.6% of all US answers, compared with 5.1% for AI Mode and 1.5% for Gemini. None of Gemini's measured links went to google.com.

Table: How the three engines linked in Findrix's US study

Measure	ChatGPT	Google AI Mode	Gemini
Links per answer (approx.)	25	3	not reported
Answers citing Reddit	35.6%	5.1%	1.5%
Product answers linking to Walmart	45.1%	not reported	not reported
Product answers linking to Amazon	6.1%	not reported	not reported
Links to google.com	not reported	46.5% of product answers link only to Google	none

Source: Findrix study as reported by Search Engine Land, October 29, 2026.

Findrix Radar adds the most-cited domains for ChatGPT across all 35 US markets. The top of the list reads like a beauty aisle: Ulta Beauty's site first, then Reddit, Walmart and Sephora. Retailers and a few health and editorial sites make up most of the rest.

Show the data

Domain	Citations
ulta.com	15397
reddit.com	14013
walmart.com	9921
sephora.com	5920
nih.gov	5215
google.com	4820
nordstrom.com	4614
target.com	4433
allure.com	3713
wsj.com	3082

Source: Findrix Radar, United States. ulta.com was 3.3% of ChatGPT's US citations and reddit.com 3.0%. Reddit was cited in all 35 markets, ulta.com in 6.

The concentration is the point. Ulta Beauty's domain was cited in only 6 of the 35 markets and still topped the list, which tells you how heavily the beauty markets lean on it. Reddit appeared in every market. For a beauty brand, the retailer product page and the Reddit thread are often the pages ChatGPT reads before it names you.

Which consumer categories are open to new brands

Findrix also reports how much of ChatGPT's brand mentions the top five brands take in each market. A low share means the leaders are weak and a newer brand has more room to be named. A high share means the answer is close to settled.

Show the data

Market	Value
Haircare	20%
Makeup and cosmetics	26%
Sports nutrition	26%
Sun care	30%
Swimwear	32%
Facial skincare	33%
Activewear	34%
Oral care	35%
Jewelry	44%
Pet food	48%

Source: Findrix Radar, United States. For comparison, residential conveyancing services was the most concentrated market at 87%.

Haircare is the most open of all 35 markets. Pet food is the most concentrated of the consumer goods markets, nearly half of mentions going to five brands. For a challenger in haircare, makeup or sports nutrition, these numbers say the field is wide and getting named is realistic. For a challenger in pet food, it says the incumbents already own much of the answer and the work is harder.

These shares come from ChatGPT only, so they do not tell you how open AI Mode is. Given how much AI Mode routes to Google Shopping, we would expect its brand mix to follow the Shopping listings closely, but Findrix does not publish that split.

How concentrated each engine's sources are

The last Radar table is about sources rather than brands. On AI Mode, the single most cited domain took 29.5% of all citations, and 53 domains accounted for half. ChatGPT needed 324 of 52,463 domains to reach half its citations, and Gemini needed 326 of 10,907.

Table: Share of each engine's US citations held by its top domains

Top domains	ChatGPT	Gemini	Google AI Mode
Top 1	3.3%	3.7%	29.5%
Top 10	15.3%	12.9%	39.5%
Top 50	27.8%	26.8%	49.6%
Top 100	35.4%	34.5%	54.9%
Domains needed for half	324	326	53

Source: Findrix Radar, United States, measured October 14 to 17, 2026.

Findrix's own summary is a useful rule of thumb: "The steeper the curve, the shorter the list you have to be on." On AI Mode the list is short and mostly Google. On ChatGPT and Gemini it is long, and a single brand page is one small source among hundreds.

What we would do with this

We would split the work by engine rather than chase one AI visibility number.

For Google AI Mode, treat Google Merchant Center as a product page. Titles, sizes, scents, ingredients, ratings, price and availability in your feed are what an AI Mode answer is most likely to show. We covered the attributes worth filling in our Merchant Center guide, and the business case in our think piece on why AI Mode makes free listings worth more. If you sell only through retailers, check that the retailers' feeds carry the same facts as your label.

For ChatGPT, follow the domains it cites. In beauty that means the Ulta Beauty, Sephora, Walmart and Target listings, plus the Reddit threads where your category gets discussed. Complete, consistent retailer listings do more here than another page on your own site.

For Gemini, the long tail matters. Half of its citations are spread over more than 300 domains, so editorial reviews, how-to guides and expert sources carry weight. Our earlier look at what 48,527 AI-cited pages have in common is the best guide we have to the pages that win there.

Whatever the engine, measure answers rather than referrals. A product answer that shows your item on Google Shopping sends no visit to your site and still sells your product. Some teams use a tracker for this, such as XENA Intelligence, which scores products by how they show up when assistants are asked shopper questions and keeps the answers per SKU in SearchPanel. A manual monthly check across the three engines, with the same twenty questions each time, tells you most of what you need.

Caveats worth keeping

Findrix sells AI visibility measurement, so it has an interest in the finding that AI answers matter. The study covers one country, four days, and one run per question. It counts links shown, not clicks. And the category results for AI Mode come from Search Engine Land's summary; Findrix did not publish the full category table for that metric on its Radar page.

Other research points the same way on engine differences. Vercite's October study of 65,746 answers found AI Mode made a top pick in 0.6% of answers that named a brand, against 10.0% for ChatGPT. Two different companies, two different methods, the same picture: the engines do not agree on how to answer a shopper, so a brand has to show up in each of them on its own terms.