AI shopping answers often disagree on which brands belong.
This benchmark looks at one part of product discovery: the brands named in AI shopping answers. We asked 15 real shopper questions across consumer tech, supplements, apparel & footwear, food & drink on ChatGPT, Gemini, Google AI Mode, kept every completed answer, and recorded which engine named which brand.
19 Aug 2026/3 engines/45 completed answers/153 brand namings
71% of brand namings came from a single engine.
153 namings, by engines naming- One engine only
- 10871%
- Two engines
- 2818%
- All three engines
- 1711%
The engines do not answer alike.
Across the same 15 questions, ChatGPT put 3.0× as many brands in front of a shopper per answer as Gemini. Where an engine answered and named nothing at all, that is recorded too.
| Engine | Answers | Brands named | Per answer | Named by it alone | Answered, named nothing |
|---|---|---|---|---|---|
| ChatGPT | 15 | 125 | 8.3 | 84 (67%) | 0 |
| Gemini | 15 | 41 | 2.7 | 9 (22%) | 2 |
| Google AI Mode | 15 | 49 | 3.3 | 15 (31%) | 3 |
One row per engine over the same 15 questions, 19 Aug 2026. A naming is one brand in one answer; “named by it alone” counts the brands no other engine repeated on that question. Volume is not quality, and this is not a ranking of the engines.
Every question, and which engine named what.
NamedNamed by all threeNot named
“What are the best noise cancelling headphones under $200?”
4 brands named in total. 3 of them by only one engine.
“Which robot vacuum should I buy for pet hair?”
7 brands named in total. 2 of them by only one engine.
“What is a good portable monitor for travel?”
8 brands named in total. 5 of them by only one engine.
“Best budget mechanical keyboard for typing?”
10 brands named in total. 9 of them by only one engine.
“What is the best creatine supplement to buy?”
11 brands named in total. 7 of them by only one engine.
“Which magnesium supplement is best for sleep?”
9 brands named in total. 6 of them by only one engine.
“Best protein powder for women over 40?”
12 brands named in total. 8 of them by only one engine.
“What electrolyte drink is best for runners?”
11 brands named in total. 7 of them by only one engine.
“Which merino wool base layer should I buy for winter hiking?”
8 brands named in total. 6 of them by only one engine.
“Best waterproof rain jacket for commuting?”
10 brands named in total. 7 of them by only one engine.
“What are good everyday sneakers for wide feet?”
9 brands named in total. 4 of them by only one engine.
“What is the best olive oil to buy for cooking?”
13 brands named in total. 11 of them by only one engine.
“Which coffee beans should I buy for espresso at home?”
13 brands named in total. 11 of them by only one engine.
“What is a good protein bar that actually tastes good?”
13 brands named in total. 11 of them by only one engine.
“Best hot sauce gift set?”
15 brands named in total. 11 of them by only one engine.
Columns run in the same order on every figure: ChatGPT, Gemini, AI Mode. The number under an engine is how many brands it named for that question, so a column of rules under a 0 is an engine that answered and named nothing. An engine that returned no completed answer is left out of the question entirely rather than shown as a zero.
How every figure on this page is produced.
The same discovery check shown on the Beseam homepage, pointed at a public category instead of a customer store. The date and denominator stay attached so the result can be interpreted in context.
- The question
- A question a real shopper would type, asked verbatim on every engine. Published exactly as it was asked: never rewritten afterwards to fit the result.
- The engines
- Each assistant is asked the same question and kept in its own column, so every mark on this page is attributable to the engine that made it.
- The denominator
- Every count is out of the engines that actually answered, and that number is printed on the figure. Nothing is averaged across questions or categories.
- The date
- Answers move. Each figure carries the date it was run and is never quietly refreshed underneath the same numbers.
- Whose data this is
- Public brands, observed from the outside. No customer store, no private catalog, and no claim about why any brand was or was not named.
- What is not claimed
- Being named is not a ranking and not an endorsement, and a brand that went unnamed is not judged here: only recorded as absent from that answer.