Between 9 and 31 August 2026 Productrise ran 100,000+ shopping queries across the US and UK and compared the products Google AI Mode recommended against the products that appeared in the standard 'Popular products' carousel on the same results page. The finding that caught the industry's attention was not that AI Mode shows different products, which most retailers expected, but how rarely the two surfaces agree and how differently they behave when they do overlap: on average, just 1.28% of products matched between the two surfaces each day, and when the same product did appear in both places, 49.6% of the time the first-listed seller differed and 38.1% of the time the first-listed price differed. When AI Mode's price was higher, the gap averaged 21.6% across all matched products, and in two-thirds of cases where prices differed, AI Mode showed the more expensive option.
For a retailer that has spent the past decade optimising to appear in Google Shopping results at the lowest available price, that behaviour reads as a different game with different rules, because the tactic that wins a Shopping carousel placement, being the cheapest seller of a popular SKU, appears to carry far less weight in AI Mode. The argument of this piece is that AI Mode is selecting for trust signals and source authority rather than for price or popularity, and that the work to appear in its recommendations looks more like earning citations in AI answers than like winning a product feed auction. The data below explains why, and what a team should test if AI Mode traffic is starting to matter to revenue.
AI Mode and Shopping carousel barely overlap
The first-order finding is how little the two surfaces have in common. Productrise measured over 2 million product listings across the 23-day period and found that AI Mode's recommendations matched the Popular products carousel an average of 1.28% of the time, which is to say that on 98.72% of checks the two surfaces were showing entirely different products for the same query. AI Mode shows an average of 3.9 products per answer while the carousel shows 27.8, so some of the divergence is down to AI Mode being a narrower surface, but even accounting for that, the overlap is far smaller than most retailers assumed before the data landed.
| Metric | AI Mode | Popular products carousel | Match rate |
|---|---|---|---|
| Average products shown | 3.9 | 27.8 | 1.28% daily overlap |
| Seller differs (when product matches) | - | - | 49.6% of matched cases |
| Price differs (when product matches) | - | - | 38.1% of matched cases |
| AI Mode price higher (when prices differ) | - | - | 68.4% of differing cases |
The interpretation that Hugo Huijer, Productrise's founder, offered to Search Engine Journal was that "the cheapest price is less of a factor in AI Mode", which the data supports: when AI Mode and the carousel did show the same product, and the prices differed, 68.4% of the time AI Mode listed the higher price, with a median difference of 22.2% (and an average of 88.5% when outliers are included). When AI Mode was cheaper, the median gap was just 7.8%, so the surface leans towards more expensive options when it deviates from the carousel baseline.
Google's response to Futurism, when asked about the discrepancy, was that "all shopping results on Google Search, including AI Mode and the search results page, are powered by the same data source: our Shopping Graph", which is factually accurate and also sidesteps the question, because the issue is not whether the data source is the same but how the two surfaces weight the signals inside that data. AI Mode and the carousel can both read from the Shopping Graph and still produce materially different recommendations if one prioritises price and popularity while the other prioritises source authority and review sentiment, which is what the Productrise data suggests is happening.
Why AI Mode behaves differently from Shopping results
The behaviour makes sense when you consider what AI Mode is built to do, which is answer a question rather than list options ranked by commercial relevance. When a user asks "which running shoes should I buy for flat feet", the traditional Shopping carousel returns the most popular or best-promoted products in that category, weighted by price competitiveness, seller rating, and bid. AI Mode, by contrast, is assembling a written answer that names specific products and explains why they fit the need, which requires the model to trust the sources it is reading and to present options it believes will satisfy the query rather than options that simply bid the highest or cost the least.
That shifts the optimisation work from winning a product feed auction to earning mentions in the sources AI Mode reads when building its answer, which in practice means review sites, community discussions, video walkthroughs, and structured product pages that the model can lift claims from. Ahrefs' wider analysis of what drives AI visibility found that third-party mentions and video content correlate far more strongly with appearing in AI answers than any on-page product detail does, and Semrush's data shows Reddit alone accounts for 40.1% of AI citations across engines, the single most-cited source. The implication for a retailer is uncomfortable but clear: if the goal is to appear in AI Mode recommendations, the highest-leverage work is not tuning your product feed but getting your products discussed in the places AI Mode trusts.
The pricing behaviour also makes sense in that frame, because a model assembling an answer based on trusted reviews and community consensus will surface the products those sources recommend, regardless of whether they are the cheapest option available. If a running shoe subreddit consistently recommends a £140 model over a £90 alternative, and AI Mode weights Reddit heavily when building its answer, the £140 shoe will appear even though the Shopping carousel would rank the £90 option higher. That is not a bug, it is the model optimising for a different signal, which is trust rather than price.
Want to get recommended by AI?
Check your AI search visibility, then let the Honeyb agent write, fix, and earn what gets you recommended. Free to start.
What retailers should test and optimise for
If AI Mode traffic is starting to matter to your revenue, the first move is to audit which products the model recommends for your category's buyer-intent queries and compare that list to what you expected based on your Shopping performance. If the products AI Mode names are the ones you rank for in Shopping, the two surfaces are aligned and you can treat them as a single channel. If AI Mode is naming different products, or naming competitors you outrank in Shopping, you are looking at a gap between what the Shopping Graph knows about your inventory and what the sources AI Mode reads say about your products, and the fix is to close that gap by earning mentions rather than by tuning the feed.
The second move is to find out which sources AI Mode is reading when it builds answers for your queries. Google AI Mode does not show citations the way Perplexity does, so you cannot read the source list directly off the answer, but you can infer it by looking at which review sites, comparison pages, and community threads rank for your category and by running queries in Perplexity (which does show sources) as a proxy for the kind of content AI engines trust. The sources that appear again and again across related prompts are your target list, and the work is to get your products mentioned, reviewed, or recommended in those places.
| What to audit | How to check it | What it tells you |
|---|---|---|
| Which products AI Mode recommends for your category | Run buyer-intent queries and log the product names across a dozen runs | Whether AI Mode sees your top SKUs or different ones |
| How often your brand appears vs. competitors | Measure recommend rate: the share of runs in which you are named | Your baseline visibility before you start optimising |
| Which sources the model trusts | Check which review sites, Reddit threads, and videos rank for your category | Where to earn mentions that will shift the AI answer |
The third move is to structure your product pages in a way that makes them easier for AI Mode to lift claims from, because even if the model is weighting trusted sources more heavily than your own page, it will still read your page if it ranks and it will use structured data, tables, and direct answers when they are available. A page that opens with a clear answer to a buyer question ("These running shoes are built for flat feet and offer arch support through a dual-density midsole") and follows with a comparison table is far more liftable than the same information buried in three paragraphs of marketing copy. That tactic is not unique to AI Mode, the same structure helps you appear in traditional featured snippets and in other AI engines, but it is table stakes if you want the model to use your page as a source rather than only reading the reviews.
The pricing data suggests you should also stop assuming that being the cheapest seller will win you the AI Mode placement, because in two-thirds of the cases where AI Mode showed a different price to the Shopping carousel, it picked the more expensive option. That does not mean pricing does not matter, but it does mean that price alone is no longer sufficient to win the recommendation, and a product with stronger review sentiment or more community mentions will beat a cheaper alternative even when both are in the Shopping Graph.
What this means for the future of AI-driven shopping
The Productrise study is a snapshot of three weeks in August 2026, but it points to a structural shift in how shopping recommendations are being generated and what that means for retailers who treat Google as a revenue channel. If AI Mode continues to grow as a share of search traffic, and if its behaviour continues to diverge from the Shopping carousel, then the optimisation playbook that worked for the past decade, bidding competitively and undercutting on price, will move a smaller share of the answer than the work of earning trusted mentions and structuring product content to be cited.
That is not a comfortable message for a retailer whose entire acquisition strategy is built around product feed optimisation, but it is a realistic one, because the engines assembling answers from trusted sources rather than ranking by bid or price is not a temporary experiment, it is the direction all of the big surfaces are moving. ChatGPT Shopping does not show a carousel at all, it names products in a written answer and explains why they fit the query. Perplexity does the same and prints its sources so you can audit them. Google AI Mode is adopting the same behaviour, and the Shopping carousel is becoming the secondary surface rather than the primary one.
For a retailer that treats AI search as a measurable channel, the move is to start tracking AI Mode visibility the same way you track Shopping performance: as a set of queries, a recommend rate for each, and a trend line that shows whether the work you are doing is shifting the answer. The free AI visibility checker gives you a baseline for a handful of prompts, and from there the discipline is to remeasure after each round of work, the same way you would for any other acquisition channel. If you wait until AI Mode is 30% of your category's search traffic, you will be optimising from behind while competitors who started measuring in August are already earning the mentions that shift the answer.
