A recommendation query is a question that asks an AI to name specific products, brands, or places ("best X for Y"). It is the query class where AI visibility is won or lost.
Informational queries ("what is a CRM?") produce explanations; recommendation queries ("which CRM should a 10-person startup use?") produce brand names. The second class carries the commercial intent, and it is the class an AI visibility index instruments.
Real recommendation queries come from real buyers: they include constraints ("free", "for agencies", "that integrates with Slack") and follow-ups. Good measurement mirrors that phrasing rather than inventing sterile benchmark questions.
Why the exact phrasing changes the answer
This is measurable rather than theoretical. Across five phrasings of one project-management question on this index — "best project management tools", "top project management tools", the same with a year appended, and the bare category — the resulting boards overlapped between 40% and 79%. Same intent, same models, materially different brands named.
So a visibility number is only as meaningful as the question behind it. A vendor reporting a score without publishing its prompt set is reporting an artifact of phrasing choices you cannot inspect.
Constraints are what make a query commercial
"Best CRM" and "best CRM for a 10-person startup that lives in Slack" are different questions with different winners. The constraints buyers actually attach — team size, budget, existing stack, industry — are what turn a browsing question into a shortlist.
Measurement that strips those constraints back to a bare head term is easier to run and less useful, because it is no longer the question anyone asks.