What AI recommends to software buyers

Each question is asked once per engine, in English, as from the United States. ChatGPT and Gemini are read the way a person sees them in those apps; Perplexity through its own sonar API with web search on. Brand names are read automatically from where an answer lists what it recommends: table rows, numbered headings and names in bold at the start of a list item. That reading is not checked by hand, so treat a single answer's list as approximate and the counts across many answers as the finding. A brand counts once per answer however often it is named, and spellings of one product are merged (HubSpot and HubSpot CRM). We keep measurements only, never the answer text. 0 of 450 requests failed and are left out. Measured on 2026-11-01; the same questions run again on 1 December 2026. The data is free to reuse under CC BY 4.0, crediting Vantage.