How does Claude decide what to put in front of you when you ask for the best headphones, the best laptop, or the best air fryer? I wanted a real answer, not a guess, so I ran the same test I built for ChatGPT: 50 shopping prompts across 10 product categories, one fresh chat per prompt, no follow-up questions, no asking for sources.
Each category got five prompts. A broad best-of query, a budget version, and three category-specific angles built around use case, feature, or a price ceiling. I recorded the top pick, the brand behind it, any source Claude named, and whether the answer gave me a shopping card of any kind.
Claude named real products, real prices, and real tradeoffs across all 50 prompts. It also sometimes displayed clickable source citations alongside those picks, publications like RTINGS and Consumer Reports showing up as clickable chips, not just plain text. What it never gave me, in any of the 50 responses, was a dedicated shopping card or product link. That distinction, between a source citation and a shopping card, turned out to matter more than I expected, and it runs through the rest of this piece.
Table of Contents
Nena’s Quick Verdict
Across 50 Claude shopping prompts, recommendation behavior was sharply category-dependent, the same pattern I found in ChatGPT. Sony won 4 of 5 headphone prompts, Apple won 4 of 5 laptop prompts, and CeraVe won 4 of 5 moisturizer prompts. Coffee makers and running shoes fell apart completely, producing five different winners each.
The finding I would lead with is a distinction, not a zero. Claude sometimes showed its work through clickable source citations. It never showed a shopping card. Source signaling and shopping functionality are not the same feature, and this test makes that gap visible.
How I Ran the Test
Every prompt ran in its own fresh chat, typed exactly as written, with nothing carried over from an earlier test. I logged the first response only. The ranked products, the brand behind each one, any publication or reviewer Claude referenced by name, and whether anything in the answer was a shopping card or product link.
I want to be upfront about a limit in my own method here. I recorded whether a shopping card appeared consistently across all 50 prompts. I did not consistently record whether Claude’s citation chips rendered as clickable in the interface for every single response, since my early notes were based on copied response text rather than the live UI. I have since confirmed, using the interface directly, that Claude’s source chips can be clickable. I cannot tell you how many of the 50 responses had them. I can tell you that at least some did.
This is also an August 2026 snapshot. Prices move, product generations move, and how Claude builds an answer can shift with a model update. Run these same 50 prompts in six months and some of the specific winners will change. The pattern behind them is what I am reporting, not a permanent leaderboard.
One more limit worth naming now. A handful of the “number one” picks were not labeled that way outright. Claude sometimes led with a product using phrasing like “most popular all-rounder” instead of a flat “best overall” tag. I treated first-position wording as the winner in those cases and flagged it in the data. Small ambiguity. Worth knowing before you read the tables.
Finding 1: Claude Shows Sources Alongside Some Product Recommendations
This is the finding that changed how I read the rest of the test, just not in the direction I first thought. When I asked Claude for the best robot vacuums for pet hair, it recommended the Roborock Saros Z70 and attributed that pick to RTINGS, with a clickable source chip sitting right next to the claim. Further down the same response, Consumer Reports appeared the same way, as a clickable citation tied to a specific pick.
That is a real citation, not just a named publication. Claude referenced RTINGS, Consumer Reports, Forbes, TechRadar, Tom’s Guide, CNN Underscored, GearJunkie, REI, Pack Hacker, Food Network, and BBC Good Food by name across different categories, and at least some of those references came with a working link a reader could click through to check.
Here is where I want to be careful. A visible citation tells you where a claim is attributed. It does not tell you that Claude checked that source in real time for this exact answer, and I am not claiming that. What I can say is that the response I saw included a clickable path from a specific recommendation back to a specific publication, and that changes the trust picture from what I originally reported.
Source citation is not the same thing as a shopping card, though. Across all 50 responses, I did not see a single dedicated product card, price box, or buy-style link. Claude will point you toward a source. It did not, in this test, hand you anything that functioned like a storefront widget.
Finding 2: Some Categories Are Basically Owned by One Brand
Four categories showed strong, repeatable concentration around one name. The table below is the cleanest summary I have.
| Category | Brand-level concentration |
|---|---|
| Laptops | Apple, 4 of 5 |
| Wireless headphones | Sony, 4 of 5 |
| Moisturizers | CeraVe, 4 of 5 |
| Air fryers | COSORI, 3 of 5 |
Apple won the general, college, video-editing, and under-$1,000 laptop prompts, all with some version of the MacBook Air or MacBook Pro M4 line. It only lost the budget slot, where Claude switched to the Lenovo IdeaPad Slim 3x. Apple dominated this five-prompt laptop cluster. Budget shoppers get routed somewhere else entirely.
Sony told a similar story. The WH-1000XM6 took the general, travel, and noise-cancelling prompts. Sony still won the budget prompt too, just with a different model, the WF-C510, which pushed its brand-level total to 4 of 5. Only the under-$100 prompt broke away, landing on the Anker Soundcore Space One.
CeraVe followed the same shape in skincare. Its Moisturizing Cream won the dry-skin, budget, and sensitive-skin prompts, and its AM Facial Moisturizing Lotion took the SPF prompt. Cetaphil only broke through on the strict under-$30 query. CeraVe dominated this five-prompt cluster, and the ingredient language behind it, ceramides, hyaluronic acid, fragrance-free formulas, showed up again and again in Claude’s reasoning.
Finding 3: Change the Intent, the Winner Changes
Now the opposite pattern. Coffee makers produced five different number one picks across five prompts: Technivorm Moccamaster, Mr. Coffee, Keurig, Breville, and a tied OXO and Cuisinart pick under $150. Running shoes did the same thing, splitting across ASICS, Adidas, Brooks, and HOKA, with Brooks winning twice.
That is a real split in how these two category types behave. Laptops and moisturizers reward one brand almost no matter how you phrase the question. Coffee makers and running shoes reward nobody consistently. Intent decides the winner more than the category does, and in a fragmented cluster, there is no single query worth chasing.
Robot vacuums sat in between. Dreame won the general and under-$500 prompts, Roborock took pet hair and hardwood floors, and Eufy won budget. Three brands, five wins, no single owner. If your product lives in a category shaped like this, optimizing for one best-of phrase is close to pointless. Each intent has its own winner, and each one needs its own plan.
Finding 4: Budget and a Price Ceiling Are Not the Same Question
This is the finding I did not expect going in. In the smartwatch cluster, Claude treated “budget” and an explicit price cap as two different questions with two different answers.
| Prompt intent | Winner | Approximate price |
|---|---|---|
| General | Apple Watch Series 11 | Full price |
| Budget | Amazfit Active 2 | $99–130 |
| Fitness tracking | Apple Watch Series 11 | Full price |
| Android compatibility | Samsung Galaxy Watch 8 | Mid-range |
| Under $300 | Apple Watch SE (3rd gen) | About $249 |
“Budget” pulled Claude all the way down to a $99 to $130 Amazfit. “Under $300” pulled it up near the ceiling, landing on a roughly $249 Apple Watch SE. Same shopper intent on the surface. Very different products underneath.
Travel backpacks moved the other direction. The budget prompt and the under-$150 prompt both converged on the same product, the Osprey Farpoint 40. So this is not a rule that holds everywhere. It is a reminder that “budget” is vague, a price cap is precise, and Claude does not always treat them the same way. If you are building content or product data around a price angle, that distinction is worth testing category by category, not assuming.
Finding 5: Sources Get Named. Shopping Cards Don’t Show Up.
The citation picture from Finding 1 holds here too. In office chairs, Claude invoked Forbes, TechRadar, and Tom’s Guide by name while walking through Herman Miller’s Sayl and Aeron lines. In robot vacuums, it named RTINGS, Consumer Reports, and Vacuum Wars while comparing obstacle avoidance and mopping performance. Some of these, based on what I have since confirmed in the live interface, came through as clickable chips rather than plain text.
The part that stayed constant is the shopping layer. Across all 50 responses, I never saw a product card, an image-plus-price box, or anything that functioned like a buy button. Claude will tell you what it thinks, and it will sometimes show you where that opinion came from. It will not, at least in this test, hand you a storefront-style widget to act on.
That matters differently depending on who you are. As a shopper, a clickable citation at least gives you a path to verify a claim yourself. As a publisher, being named and linked inside an answer like this is a real visibility signal, worth more than a plain-text mention. Whether that translates into meaningful traffic is a separate question this test cannot answer, since I did not track click-through behavior, only what appeared in the response.
The Travel Backpack Cluster, in Full
This category gave me the cleanest side-by-side view of the recommendation pattern, so it is worth laying out prompt by prompt.
| Test | Intent | Number one pick | Shopping cards |
|---|---|---|---|
| 46 | General | Cotopaxi Allpa 35L | 0 |
| 47 | Budget | Osprey Farpoint 40 | 0 |
| 48 | Carry-on | Nomatic Travel Bag 40L (inferred) | 0 |
| 49 | International travel | Cotopaxi Allpa 35L | 0 |
| 50 | Under $150 | Osprey Farpoint 40 | 0 |
Cotopaxi took the broad and international prompts. Osprey took both price-sensitive prompts, budget and under $150, with the exact same Farpoint 40 winning both times. Nomatic broke through only on carry-on, and even then Claude never used a flat “best overall” label, just first-position phrasing that read like a winner.
Compare that to what I found running this same cluster through ChatGPT. ChatGPT gave Osprey a clean five-for-five sweep, cited Pack Hacker in every single prompt, and showed shopping cards on three of the five answers. Claude showed no shopping cards anywhere in this cluster. Same category, same style of question, a genuinely different interface underneath it on the shopping side, whatever the citation picture looked like prompt by prompt.
What Makes a Brand More Likely to Be Recommended by Claude?
I want to be careful here. Fifty prompts run once is a strong pattern, not proof of how the system works underneath. Even so, a few things looked worth acting on.
Owning a specific use case beats chasing a generic best-of slot. HOKA did not win running shoes broadly. It won long-distance, specifically, tied to cushioning and shock absorption. Roborock did not win robot vacuums broadly. It won hardwood floors and pet hair, tied to specific mopping and suction claims. A narrow, well-earned reason to recommend something looks more durable than trying to be everyone’s general answer.
Category concentration is real, and it is worth checking before you invest anywhere. If a category already has an Apple, a Sony, or a CeraVe sitting at 4 of 5, competing head-on for the general query is a hard, possibly wasted fight. Competing for the budget slot, the compatibility slot, or the use-case slot that the leader does not own looks like the more winnable path.
Being cited and being shoppable are two different wins. Claude referenced real publications repeatedly throughout the test, and at least some of those mentions came with a working link. None of them came with a shopping card. If your AI visibility strategy is built around citations alone, this test suggests that gets you attribution, not necessarily a storefront-style placement, at least for Claude, at least right now.
The Limits of This Test
I ran each prompt once, in a fresh chat, during one testing window in August 2026. Repeat the same 50 prompts next month and some of these rankings will move, especially anywhere price or product generation drives the answer. Fifty prompts is a real sample for manual testing. It is not a statistically complete slice of every commercial question a shopper might type.
I also did not ask Claude to explain its reasoning or reveal what sat behind an answer. Everything here comes from what Claude actually showed me, not from the model describing its own process.
One methodology gap is worth repeating here directly. I tracked shopping cards consistently across all 50 prompts, and that number held at zero throughout. I did not track the presence of clickable citation chips with the same consistency, since part of my original logging relied on copied text rather than the live interface. I now know clickable citations can and do appear. I do not have a reliable count of how often.
Related Reading
How Does ChatGPT Decide What Products to Recommend?
Do News Websites Block AI Crawlers? I Tested 13 Major Publishers
How I Found My Own AI Search Visibility Gap
How to Check AI Visibility for Free (With Real Data From My Site)
Frequently Asked Questions
Does Claude show shopping cards or product links when recommending products?
No. Across all 50 prompts in this test, Claude gave 0 shopping cards and 0 product-card style links. It named specific products, prices, and tradeoffs, and sometimes cited sources, but nothing functioned like a storefront widget.
Does Claude cite its sources when it recommends a product?
Sometimes, and those citations can be clickable. In one example, asking for the best robot vacuums for pet hair, Claude recommended the Roborock Saros Z70 with a clickable RTINGS citation, and a Consumer Reports citation appeared later in the same response. I did not consistently record how many of the 50 responses included clickable citations, so I cannot give a frequency here, only that it happens.
Does Claude always recommend the same brand for a given category?
It depends heavily on the category. Laptops, headphones, and moisturizers each had one brand win 4 of 5 prompts. Coffee makers and running shoes went the other way, producing five different winners across five closely related prompts.
Is “budget” the same as a specific price cap in Claude’s answers?
Not always. The smartwatch cluster showed a clear gap: “budget” led to a roughly $99 to $130 Amazfit, while “under $300” led to an Apple Watch SE closer to the $249 ceiling. Travel backpacks, by contrast, gave the same winner for both the budget and under-$150 prompts. The behavior is not consistent across categories.
Is this benchmark repeatable?
The prompts and method are repeatable. The results are not guaranteed to match, since Claude’s answers, pricing, and product availability all shift over time. Treat this as a snapshot from one testing period in August 2026, not a fixed ranking.
