Named is not recommended: two new studies on where AI brand picks come from

Almost every AI visibility dashboard starts from the same number: how often your brand is mentioned. It is easy to count and easy to chart. Two studies published in October suggest it is also the least useful number on the page.

Vercite, a visibility tracking company, read 152,195 brand mentions in 65,746 answers from five AI engines and sorted each one by what the answer did with the brand. Onely, a search agency, asked five engines the same 31 questions and classified every page each engine cited. Read together, they explain why a brand can be "visible" in AI and still never be the one an assistant suggests.

Neither study is about groceries or beauty, and we come back to that below. But the mechanics they describe apply to any brand that wants to be recommended.

Most mentions are not recommendations

Vercite sorted every mention into five types: a top pick ("if you just want one, go with Brand A"), an endorsement ("best for small teams"), a plain listing among options, a passing mention, and a brand that appears only in a source link. A recommendation, in its terms, is a top pick or an endorsement.

In a sample of 481 mentions reviewed one by one, 1.5% were top picks and 16% were endorsements. Most brands, 59%, were simply listed. When Vercite classified all 152,195 mentions automatically, it got 1.0% top picks and 15.5% endorsements.

At the level of whole answers, 53% named at least one brand, rising to 68% when the prompt asked for a recommendation. But only about one in 20 answers that named a brand made a top pick, and Vercite estimates about six in ten recommended no brand at all.

That is the gap the mention count hides. We made the same point from a different angle in famous is not the same as recommended: being known to the model and being chosen by it are different outcomes.

The engines recommend very differently

To compare engines fairly, Vercite used 339 prompts that all five answered. The spread is wide.

Show the data

Engine	Value
ChatGPT	10%
Google AI Overview	6.5%
Perplexity	6.1%
Gemini	1.9%
Google AI Mode	0.6%

Source: Vercite, How often does AI actually recommend a brand?. Automated count; Vercite's checks suggest the true top-pick share is slightly higher.

Table: How each engine treats the brands it names

Engine	Answers naming a brand	Brands per answer	Answers with a top pick	Answers that only endorse	Mentions endorsed or picked
ChatGPT	74%	4.4	10.0%	38%	24%
Perplexity	58%	3.9	6.1%	32%	24%
Google AI Overview	61%	3.8	6.5%	22%	16%
Google AI Mode	56%	4.2	0.6%	19%	9%
Gemini	69%	4.5	1.9%	32%	16%

Source: Vercite, October 13, 2026. Shares of answers cover answers that name at least one brand.

ChatGPT is the engine most willing to choose. Google AI Mode almost never does: 80% of its brand-naming answers recommend no brand, and only 9% of the brands it names are endorsed or picked. That fits what Findrix found for US product questions this week, where AI Mode kept 46.5% of product answers on Google's own pages and leaned on Shopping listings rather than editorial verdicts.

The engines also disagree with each other. When two engines answered the same prompt on the same day, they named the same brand first only 28% of the time. A good showing in one says little about the next, which is one more reason we argued against tracking a single AI rank.

Position and citations go together with being picked

Two patterns in Vercite's data are directly useful.

First, order matters. 91% of top picks were the first brand named in the answer. Across all answers, the first-named brand was endorsed or picked 25% of the time and the tenth-named brand 8%.

Second, which pages are cited matters, and the brand's own site is not the strongest signal. Vercite compared brands in the same position and split them by citation:

Table: Brands named first in an answer, by what the answer cites

Citation status	Endorsed or picked	Top pick
Not cited	26%	4%
Own site cited	28%	6%
Named on another site's cited page	37%	8%
Both	42%	13%

Source: Vercite, October 13, 2026. "Named on another site's cited page" means the brand is in the title of a cited page on someone else's site.

Vercite is careful to call this a link, not proof. Engines may cite comparison and review pages because those pages rank brands, and then repeat the ranking. But the direction is consistent across all five engines, and its summary is a fair one: "a brand's own site helps an engine describe it, while other sites are more closely tied to it being recommended."

Count how often you are endorsed or picked, not only how often you are named.

The wording of the question matters as well. In Vercite's data, prompts asking for "the best" got a recommendation in 50% of answers that named a brand and a top pick in 8%. Prompts asking how, what or why almost never produced a top pick. Shopper questions in packaged goods are full of both kinds: "best dry shampoo for oily hair" invites a pick, while "how often should I wash my hair" mostly produces advice with brands in passing. A brand that wants to be picked has to show up in the first kind of question, and a brand that wants to be trusted has to be quoted accurately in the second.

What each engine reads before it decides

Onely's study looks at the same question from the source side. It asked 31 questions of the form "best agencies for X" to five engines, with web search switched on in ChatGPT, and classified every cited page by type.

Named brands' own sitesLists of the best agencies

Show the data

Engine	Named brands' own sites	Lists of the best agencies
ChatGPT	57%	22%
Perplexity	7%	70%
Gemini	14%	77%
AI Mode	9%	59%
AI Overview	4%	48%

Source: Onely, How ChatGPT Decides Which Brands to Recommend. B2B agency questions. AI Overview appeared for 18 of the 31 questions. Directories, Reddit and other pages make up the rest.

ChatGPT stands apart. 119 of the 210 pages it cited, 57%, were the websites of the agencies it named. Perplexity, Gemini and AI Mode built their answers mostly from roundup lists. Reddit was a small share of citations everywhere, 1% for ChatGPT, even though Reddit threads were 698 of the 2,521 pages ChatGPT's searches returned. ChatGPT read Reddit and did not cite it.

Onely also notes a limit on its own result. ChatGPT does not search for every question in ordinary use; it cites Semrush data that ChatGPT used web search on 34.5% of queries in February 2026, down from 46% in late 2024. When it does not search, it answers from training, where well-covered brands have the edge. Onely cites a University of Toronto study of soda prompts in which major brands got 56.3% of ChatGPT's mentions and niche brands 12.3%.

Do the two studies disagree?

At first glance, yes. Vercite finds that being named on someone else's cited page goes with more recommendations than having your own site cited. Onely finds ChatGPT cites brands' own sites most.

Look closer and they fit. Vercite's engine breakdown shows the brand's own site helps most on ChatGPT, where brands with their own site cited were endorsed or picked 24% of the time against 14% for uncited brands, and least on Perplexity. Onely's data says the same thing from the other side: ChatGPT goes to brand sites, the other engines go to lists. So the honest summary is that ChatGPT rewards a clear, quotable brand site, and the other engines reward a place on the third-party pages they read.

What this means for a CPG brand

Both studies have prompt sets far from the grocery aisle. About half of Vercite's prompts were in Swedish and leaned toward finance, real estate, hospitality, energy and business services. Onely asked only about B2B agencies, and it sells the services its findings recommend, so we use its counts and set aside its conclusions about itself. Neither study's exact rates should be read as a benchmark for snacks or skincare.

The mechanics still translate, and they suggest four habits.

1. Track endorsements, not mentions. When you review AI answers for your category, note whether your product was picked, endorsed for a need ("best for sensitive skin") or merely listed. Vercite's data says the last is the norm.
2. Give engines a "best for" to repeat. Most recommendations were tied to a need. Product pages that say plainly who the product suits, and who it does not, give ChatGPT something to quote. Our tip on being recommended for specific needs covers the page work.
3. Earn a place on the pages other engines read. For Perplexity, Gemini and Google's AI features, that means editorial reviews, buying guides and comparison articles in your category, plus complete retailer listings. Our guide to getting into cited sources is the starting point.
4. Check each engine separately. A brand that ChatGPT picks may only be listed in AI Mode. Some teams use trackers for this; XENA Intelligence, for one, scores how products show up in assistant answers, with the answers kept per product in SearchPanel. A fixed list of shopper questions, run monthly in each engine, will show the same pattern.

The common thread is that recommendation is earned per engine and per need. A mention count that rises month after month can sit on top of a top-pick rate that never moves, and only the second one sells product.