A test of AI brand recommendations across four answer engines found a wide gap between recognizing a brand and recommending it to buyers. Tony Uphoff asked ChatGPT, Gemini, Perplexity and Google AI Overviews questions prospective clients might ask, then tracked whether two businesses he launched in February appeared.
Across 59 answers, Uphoff on Media appeared in none of 12 category answers, while Uphoff Advisory appeared in one of 43. Yet every engine answering a question about Uphoff by name knew his career. He described the difference this way: “The models know who I am. They almost never recommend what I’ve built.”
The results offer a concrete example of why AI brand recommendations may need a different measurement than name recognition. The test is a snapshot, not a broad industry benchmark, and its results do not establish how often other brands appear.
Why AI Brand Recommendations Need Separate Measurement
Uphoff argued that teams can mistake an accurate answer to a branded question for evidence that their business is visible to potential buyers. His test instead used category questions, where the answer engine had to identify possible providers without being given the company name.
He cited a March 2026 G2 survey of 1,076 B2B software buyers. In that survey, 51% said they now start research with an AI chatbot more often than with Google, compared with 29% in G2’s 2025 report. Also, 69% said a chatbot led them to choose a different vendor than they had planned, and one in three bought from a vendor they had not heard of before.
Those findings describe surveyed buyers, while Uphoff’s test measured two businesses. Being retrievable by name does not show whether a business enters a buyer’s shortlist. A related look at AI Mode results discusses how answers can present information about people and companies.
Old Information Can Shape the Answer
The engines that described Uphoff still called him CEO of Pipeline360, he said, even though he now runs Uphoff Advisory. He said only one mentioned his current firm. He attributed the outdated descriptions to sources such as organization-chart aggregators, executive databases and old press releases.
Uphoff’s observation was that answer engines often rely on what the web has said most often, which may be older information. He also said a Corporate Ink study reported by eMarketer in May found 29% of marketers actively track and measure AI visibility.
Three engines accurately defined Demand-as-a-Service and cited Pipeline360 pages, but did not mention Uphoff, who helped create the strategy. He accepted that because the company owns the category language. He said models credit the clearest published explanation, not necessarily the idea’s originator.
A guide to AI search visibility covers how brands approach information retrieval and citations.
Related reading on search visibility beyond Google rankings explores the broader measurement context.
Measure the Shortlist Over Time
Uphoff plans to rerun the same prompts in January and publish the results. The source author said one week and 19 prompts provide only a snapshot because AI answers can change from run to run. The planned rerun could show whether the initial results persist, though the source does not report its outcome.
Uphoff recommends asking 10 to 15 buyer-style questions about a category, a problem and whom to hire, plus two or three branded prompts. He suggests running them across ChatGPT, Gemini, Perplexity and AI Overviews, tracking appearances and description accuracy.
He recommends repeating the exercise weekly, reviewing cited sources, updating company pages and seeking outside-directory corrections. He also advises publishing a clear explanation of a narrow category phrase, pursuing relevant roundup coverage, and connecting recommendation rates with signups, demo requests, subscribers and pipeline.
The test does not show that any single change will secure a recommendation. It does identify a measurable next step: repeat category prompts and compare the resulting shortlist and descriptions over time.
