In a recent test, Tony Uphoff asked ChatGPT, Gemini, Perplexity, and Google AI Overviews the questions his prospects ask him, then counted how many times these response engines recommended the two companies he launched in February. The results reveal a gap between being known and being recommended, which the author believes is at the heart of the problem of measuring brand visibility.
- AI engines know Tony Uphoff’s background, but they rarely recommend his companies.
- Engines often describe outdated profiles because they favor the oldest online sources.
- Buyers increasingly start their research with AI chatbots, and these often change their provider choice.
- Brands should track how often AI engines recommend them, placing this metric before rankings and sessions.
Out of 59 responses, his newsletter, Uphoff on Media, did not appear in any of the 12 responses concerning a category, and his firm, Uphoff Advisory, appeared in only one response out of 43. Yet, every engine that answered the question ‘Who is Tony Uphoff?’ knew his background. ‘The models know who I am. They almost never recommend what I’ve built,’ he summarizes.
This gap fits into a shift in purchasing habits. According to a G2 survey published in March 2026 of 1,076 B2B software buyers, 51% now start their research with an AI chatbot more often than with Google, compared to 29% in the 2025 report. For 69% of them, a chatbot led them to choose a different provider than planned, and one in three buyers worked with a provider they had never heard of.
Tony Uphoff believes marketing teams have transposed their old benchmarks to AI. ‘For two decades, the dashboard was discoverability. SEO taught all marketing teams to rank, get the click, and be found. Most companies have carried that dashboard directly over to AI.’ Asking a model your own name, getting an accurate answer, and concluding that you are covered is, in his view, a false positive.
Engines That Describe Yesterday’s Company
The second finding concerns companies that have changed their name, merged, or undergone a change in leadership. Three out of three engines still present Tony Uphoff as CEO of Pipeline360, and only one mentions the company he runs today. ‘Sources included org chart aggregators, executive databases, and old press releases,’ he explains. According to his rule, engines favor what the web says about you most often, ‘and most often usually means oldest.’
This outdated profile carries weight twice over, since the model uses it to decide whether to recommend a company, and the human uses it to verify what the model says. Yet few teams monitor this point. A study by Corporate Ink, relayed by eMarketer in May, indicates that only 29% of marketers actively track and measure their visibility in AI.
The test offers another lesson. When Tony Uphoff asked ‘What is Demand-as-a-Service?’, all three engines defined the term accurately and cited Pipeline360’s pages, without ever mentioning the man who helped develop this strategy. ‘The models credit whoever publishes the clearest explanation, not whoever had the idea first.’
Lists Built from Other Lists
Only once did Uphoff Advisory appear on a list: ChatGPT ranked it second out of four for the question ‘Who advises founders of B2B information and data businesses?’. On category questions, the models rarely cited the companies they named, but rather rankings published by Zapier, Content Allies, and others.
A May 2026 study by Maryanna Franco and João da Silva, conducted on the Friction AI platform using 14,140 queries on sportswear brands, shows the same mechanism. Nike and New Balance share the same description in Google’s Knowledge Graph, but Nike appears in about 71% of recommendation responses on athleisure, while New Balance almost never does.
The article’s author, however, nuances this test: nineteen queries over a week constitute a snapshot, and responses vary from one run to another. An appearance in one out of 43 could therefore be a floor or noise. He approves of Tony Uphoff’s project to rerun the same queries in January and would go further by tracking the measurement weekly.
A Recommendation Rate at the Top of the Dashboard
Tony Uphoff notably proposes drafting 10 to 15 queries as buyers formulate them, plus two or three brand queries, running them on ChatGPT, Gemini, Perplexity, and AI Overviews, and tracking the frequency of appearance as well as the accuracy of the description. He recommends putting this indicator first in reports, ahead of ranking and sessions.
