57% of Google AI Mode product links stay inside Google Shopping

For swimwear, Google AI Mode linked only to Google pages in 60.4% of answers, according to a Findrix study reported by Search Engine Land. Shoppers can discover products through AI Mode without visiting the brand's website.

That figure comes from a Findrix study of 55,198 AI-generated answers, reported by Search Engine Land. Across U.S. product answers, AI Mode linked only to Google's own pages 46.5% of the time. Counted link by link, 57% of the links in AI Mode product answers went to Google Shopping product pages.

For an ecommerce director, the problem is simple. Referral traffic alone won't capture all of a catalog's visibility in Google's AI search results. The more useful question is what Google displays when it shows the product.

Swimwear stays home, software leaves

The Google-only share swings widely by category. Fashion and beauty sit at the top: after swimwear come women's everyday fashion at 57% of answers, sun care at 54.1% and activewear at 53.8%. Sports nutrition is the outlier at just 7.6%, even though it sits close to activewear on many store shelves.

Outside physical products, the pattern flips. Only 5.1% of AI Mode's links for local services such as plumbers, lawyers and dentists went to Google pages. For software questions it was 1.1%. The direction is clear.

Findrix co-founder Yosif Merman puts the gap down to Google's product database, which holds prices, ratings and inventory. With that data on hand, he said, AI Mode can build product answers with fewer links to external websites. His explanation suggests the Google-only share rises with how much structured product data Google already has in a category.

The size of that database makes the explanation plausible. Google's Shopping Graph holds more than 50 billion product listings, refreshed hourly, according to AI search firm Goodie. That scale suggests why services and software sit at the bottom of the table. The database Merman describes is built around products with prices and stock levels, so a question about a dentist or a software tool has to send the shopper somewhere else.

A different contest from the carousel

A second dataset shows that a listing which wins in classic Shopping won't automatically win in AI Mode. Productrise, which sells organic product tracking for Google Shopping, tracked more than 2 million product listings across over 100,000 regular search results and AI Mode responses between August 9 and August 31, 2026, in the U.S. and U.K. Search Engine Journal walked through the method: matching searches run on the same day, products matched by Google product ID, and the first-listed offer compared on each side.

The comparison is narrower than all of search. On the regular-results side, Productrise counted only the Popular products carousel and ignored other results modules. On the AI Mode side, it counted the product listings shown inside the answer.

According to Productrise, just 1.28% of carousel products showed up in AI Mode when the same query was run on the same day. The number is calculated per day. For each day, Productrise took the count of products found in both places and divided it by the total carousel listings that had a Google product ID. It then averaged those daily figures over the study's 23 days. Even on that generous measure, a product placed in the carousel seldom made it into AI Mode.

Prices also differed. The first-listed price was different on 38.1% of matched products, and AI Mode showed the higher price in 68.4% of those cases. When AI Mode was higher, the median gap was 22.2% above the carousel price and the average was 88.5%, according to Common Thread Collective's breakdown of the study. An average that far above the median suggests a smaller group of very large gaps is pulling the figure up.

Productrise says price and seller changes often happen together. That suggests at least part of the price gap comes from AI Mode putting a different merchant in the top slot, not from the same merchant charging more. For a brand that sells through resellers and marketplaces, that matters: winning the product doesn't guarantee winning the click.

Different inventory doesn't appear to explain the split. In September 2026, Google confirmed that Shopping results on both surfaces draw from the same Shopping Graph. AI Mode uses Google's Gemini models to generate conversational answers, and Common Thread Collective reads the Productrise data as evidence that its selection logic is distinct from standard Shopping auction mechanics. Same catalog, different judge. With fewer than four slots per AI Mode query against nearly 28 in the carousel, each position is worth more.

Both studies have limits. Productrise's queries centered on products it monitors, not a random sample of searches. Findrix measured where links point, not what shoppers bought. They show where visibility sits, which is exactly the thing referral reports miss.

ChatGPT and Gemini send shoppers elsewhere

Other assistants behave very differently, even within Google. None of the Gemini links Findrix measured went to google.com. ChatGPT packed in about 25 links per answer, compared with about three for AI Mode. In ChatGPT product answers, Walmart showed up in 45.1% of cases and Amazon in 6.1%.

AI Mode sends 57% of product-answer links to Google Shopping, while none of Gemini's measured links went to google.com. That gap is a good reason to distrust any single "AI visibility" number. A blended score could average a surface that keeps shoppers inside Google Shopping with one that sends them to a big retailer's site. For a brand selling through Walmart's marketplace, ChatGPT exposure could show up in that channel's sales rather than in direct-site traffic.

Google is building checkout into the same surfaces

The links tell only part of the story, because Google keeps adding steps that happen after discovery. Shopping in AI Mode, announced at Google I/O in 2025, pulls product information, stock status, color options and prices into one view, according to Webdesk Solution. Its agentic checkout lets a shopper set a preferred color, size and budget. The system then watches for a match, such as a price drop, and can fill in purchase details, add the item to the cart and prepare checkout through Google Pay, with the shopper confirming before anything is bought.

Apparel gets an extra layer. Virtual try-on lets shoppers upload their own photo and see a simulation of how a garment drapes and fits. Those tools land in the same fashion categories where Findrix found the highest Google-only shares, which could make those answers even more self-contained.

On January 11, 2026, Google went further. Vidhya Srinivasan, VP and GM of Ads and Commerce, announced the Universal Commerce Protocol, an open standard meant to let AI agents and merchant systems work together from discovery and buying through post-purchase support. Google co-developed it with Shopify, Etsy, Wayfair, Target and Walmart, and more than 20 others endorsed it, including Best Buy, Macy's Inc., The Home Depot, Mastercard, Stripe and Visa. The same announcement covered Business Agent, which lets shoppers chat with a brand directly on Search, and Direct Offers, which let retailers show exclusive discounts in Google Ads to shoppers ready to buy.

Walmart sits on both sides of this story. It helped build Google's protocol, and it also turns up in 45.1% of ChatGPT product answers in Findrix's data. For retailers of that size, the question may be less which assistant wins and more how to be present on each surface's terms. Paid media is following the same path: Common Thread Collective notes that in 2026, Shopping ads have been appearing in carousel format inside AI-generated search results.

The product feed does the selling

When 57% of AI Mode's product links land on Google Shopping, the Merchant Center feed is the storefront AI Mode shows most often. Google's January announcement pointed the same way, adding new Merchant Center data attributes meant to help retailers get discovered in conversational commerce.

Those pieces fit the Findrix pattern. The more of the journey happens on Google's surfaces, the more the merchant's structured data does the selling. That could matter most for swimwear, women's everyday fashion, sun care and activewear, where AI Mode linked only to Google pages in 53.8% to 60.4% of answers. For those teams, attribute completeness and price and inventory accuracy in Merchant Center deserve the attention that product page copy used to get.

Common Thread Collective suggests a starting sequence for brands already running Shopping campaigns:

1. Check AI Mode impressions on their own in Google Search Console, which segments impressions by search type, to measure how often products appear there compared with the traditional carousel.
2. Audit Merchant Center data quality. AI Mode selection appears to favor products with richer data, including detailed descriptions, accurate pricing and high-quality images.
3. Rewrite titles and descriptions for conversational queries. Traditional Shopping titles are built for keyword matching, while descriptions that read naturally and answer buyer questions directly may surface more often in AI Mode.
4. Report AI Mode and traditional Shopping performance separately, so AI Mode results don't disappear into blended Shopping metrics.

For apparel, the detail goes down to the attribute. Webdesk Solution advises sellers to supply material, size, fit and color, keep pricing and inventory current, and upload images from multiple angles, warning that incomplete listings or low-quality visuals may be left out of AI Mode's visual panels. Given the Productrise price findings, a stale price in a reseller's feed is also a visibility problem.

The number to watch is sports nutrition, where only 7.6% of answers stay inside Google today. If that share climbs toward fashion's levels, it could suggest that the feed-first approach applies far beyond apparel.