We ran 50 buying prompts across 12 B2B SaaS categories

Every vendor we talk to has seen one screenshot of an answer engine naming them, and almost none of them know how often it happens. That gap is the reason this panel exists. We built a fixed set of 50 commercial-intent prompts for 12 B2B SaaS categories, ran them against five engines, and scored who got named. No prompt was rewritten to produce a better answer, and no run was discarded for being unflattering.

The headline result is not that vendors score badly. It is that the distribution has almost no middle.

Fifty prompts per category, written the way a buyer writes them rather than the way a marketer would. Roughly a third are category-entry questions ("best X for Y"), a third are comparison questions naming one incumbent, and a third are problem-first questions that never name a category at all. Prompts are frozen for the quarter so runs stay comparable 1.

• Five engines: ChatGPT, Claude, Perplexity, Microsoft Copilot and Google AI Overviews.
• Three runs per prompt per engine, on different days, from clean sessions.
• A vendor counts as named only if the answer text names it. A link in a source tray does not count.
• Null results are recorded. A category where nobody is named is a finding, not a failed run.

Sorting every tracked vendor by how many of the 50 prompts named them produces a curve that is close to vertical at both ends and nearly empty between them.

Seventy-three per cent of tracked vendors are named five times or fewer. Those companies have marketing teams, content calendars and SEO retainers. The work is happening. It is landing somewhere the answer does not read from.

The two categories with the widest spread between leader and median were also the two with the most active independent review coverage. The two tightest were categories where the buying conversation happens mostly in private communities that models cannot read. Where the public record is thick, position is winnable and defensible. Where it is thin, everybody scores near zero and the leader's position is fragile 2.

FigureFig. 01

If you are at zero, the first question is not what to publish. It is which public surfaces in your category the engines are actually reading, and whether your name appears on any of them at all. That is a research problem before it is a content problem.

If you are somewhere between one and fifteen, you already have a position. The work is finding the prompts where you are named and the adjacent ones where you are not, then closing the specific gap rather than publishing more of what already worked.