For one sensitive-skin cleanser question, Perplexity usually named more product brands than ChatGPT. It still began with the same brand.
We reviewed saved responses to “what's the best facial cleanser for sensitive skin.” ChatGPT named two product families in 28 of its 32 answers. Perplexity named four in 23 of its 28. Both named Vanicream first in every answer.
Those counts describe literal names, not the number of recommendations. We excluded sources such as the American Academy of Dermatology and counted multiple cleanser variants from one brand only once. “First named” is also separate from an explicit best-overall choice, even when they coincide.
The difference was smaller for “best accounting software for small business.” ChatGPT named four or five product families per answer; Perplexity named five in all 12. Both had a median of five and named QuickBooks first every time. QuickBooks and QuickBooks Online were counted as one family, as were Wave and Wave Accounting.
Google AI Overviews had a median of three named families for the cleanser question and four for accounting. Its responses repeated heavily: only four distinct cleanser texts and two distinct accounting texts appeared. Repeated wording limits how much weight to put on those observation counts.
The pilot does not establish that one assistant always offers a broader shortlist, or that more names make an answer better. It shows that answer breadth can differ for a particular question while the first named brand stays the same. Recommendation scoring needs its own review; literal name counts cannot substitute for it.
These are the same six runs and 128 saved answers used in the earlier pilot articles, not a new study. Claude was not measured, so the original placeholder’s Claude comparison remains untested.