We sorted 18,618 buyer answers by the shape of the question
The sample is every answer Proofsource collected from October 4 to 26, 2026, for an English buyer question that doesn't name the brand, on ChatGPT, Perplexity, Claude and Google AI Overviews: 18,618 answers to 2,346 questions, for 119 brands in 80 workspaces.
We grouped questions by their wording. How-to questions start with "how" ("How can a small contractor quantify drywall from plans?"). Best-of questions ask for the best or top option. The others ask for alternatives to a named product, compare two named products, or ask which product does a specific job.
How-to answers named the brand 4.5% of the time, best-of answers 13.6%
Share of answers that named the tracked brand, with 95% intervals:
- Best or top: 13.6% (679 of 5,001; 12.7% to 14.6%).
- Which product does a specific job: 13.5% (535 of 3,975).
- Alternatives to a named product: 13.1% (539 of 4,113).
- How-to: 4.5% (63 of 1,401; 3.5% to 5.7%).
- One named product against another: 2.7% (64 of 2,331).
The gap held on all four engines and inside the same brands
How-to against best-of, by engine: ChatGPT 7.1% against 16.3%, Claude 4.3% against 12.1%, Google AI Overviews 3.4% against 12.1%, Perplexity 3.2% against 13.7%.
Counting questions instead of answers tells the same story. 21 of 172 how-to questions (12.2%) named the brand at least once in the period, against 192 of 628 best-of questions (30.6%).
A category mix effect would show up here, so we compared within brands. 71 brands tracked both kinds. The best-of rate was higher for 30 of them and the how-to rate for 6. The other 35 were never named on either.
Most how-to answers named nobody, and some used your page anyway
62.6% of how-to answers named none of the brands or competitors being tracked, against 34.5% of best-of answers. A how-to answer explains a method. A product shows up only when the method needs one.
The brand's own site was cited in 80 how-to answers (5.7%), more often than the brand was named (63). In 41 of those 80, the engine cited the brand's page and didn't name the brand. On a how-to question, being the page the engine reads is a realistic win, and a mention rate scores it as zero.
How-to questions are still a large share of what people ask
This isn't a case for dropping them. In the working paper "How People Use ChatGPT", which classified about 1.1 million sampled consumer ChatGPT conversations from May 2024 to June 2025, how-to advice made up 8.5% of messages and questions about products to buy 2.1%.
Many of your buyers start with how. They just get a method, and maybe a source, before they get a shortlist.
What to do with your own prompt set
- Report how-to questions and best-of questions as separate rates. A set that is 40% how-to would pull a 13.6% rate down to about 10% with nothing else changing.
- On how-to questions, count citations of your own pages next to mentions. That is where your guides earn their keep.
- Track a head-to-head question between two other products only if your buyers really ask it. You'll rarely be named in the answer.
- When you add or remove questions, check the mix before you read the trend.
How Proofsource handles it
Proofsource tracks each phrasing as its own question, grouped into topics, on every engine in the plan. For each answer it records the brands named and the pages cited, so a how-to answer that read your page without naming you shows up as a citation rather than a blank.
Where this could be wrong
Shapes come from the wording, so some questions land in the wrong group. "How do I pick the best" counts as best-of here, and a how-to question can be commercial underneath.
"Named nobody" means none of the brands and competitors our users track. An answer may have named a product nobody tracks.
This is observational. How-to questions may differ from best-of questions in ways the within-brand check doesn't catch, and answers to the same question aren't independent, so the true uncertainty is wider than the intervals shown.
The sample is 23 days, English questions asked from the US, and the categories our users chose to track. The paper's figures cover consumer ChatGPT plans to mid 2025 and use its own topic labels, not ours.
