Branded vs unbranded questions

The distinction is mechanical before it is philosophical. Ask an assistant whether Acme handles recurring invoices well and the answer will contain the word Acme almost by necessity, since the question put it there. Ask the same assistant for the best invoicing tool for a freelance designer and Acme appears only if the engine's retrieved sources and internal associations actually surface it. The first mention is produced by the question; the second is produced by the market's information environment. A visibility score is supposed to measure the second thing, and every branded question mixed into its denominator substitutes the first.

The inflation this causes is large and quiet. A question set that is one third branded hands a brand a guaranteed mention on a third of its runs before any measurement begins, and the resulting score can look respectable for a brand that no assistant ever volunteers. The failure is common in both vendor dashboards and internal spreadsheets, usually without disclosure, sometimes without awareness. The diagnostic is simple: a strong reported score alongside unbranded spot checks that keep returning competitor lists means the denominator is contaminated.

The stakes come from where discovery now happens. G2 found in 2026 that 51 percent of software buyers start research with an AI chatbot more often than with Google, and that 33 percent bought from a brand they had never heard of before an AI named it. Unprompted recommendation is the event those numbers describe, and unbranded questions are the only instrument that measures it.

Unbranded questions measure discovery and belong in the headline score. A good unbranded set reads like the real questions buyers ask assistants: specific about the job, the buyer and the constraints, phrased conversationally, and spread across the funnel from early research through comparison to final selection. It is held fixed and versioned so scores stay comparable, and it is sized to the category, covering the questions that carry commercial weight rather than every phrasing imaginable.

Branded questions measure accuracy and sentiment, and they earn their keep outside the headline number. Once a buyer knows a brand's name, the questions change shape: whether the product is good for a use case, what it costs, how it compares to a named rival, what its weaknesses are. Tracking these reveals what engines actually say when the brand is in play, which surfaces a different failure class: outdated pricing, features attributed wrongly, confusion with similarly named companies, or a rival consistently framed as the safer choice in direct comparisons.

The two sets together form a complete instrument. Unbranded tracking answers whether the brand gets into conversations; branded tracking answers what happens once it is there. A program that runs only branded questions mistakes recognition for discovery. A program that runs only unbranded questions can win the shortlist while engines quietly misdescribe its pricing to every buyer who follows up. Reachroller runs both by design and keeps them apart where it matters, excluding branded questions from the headline visibility score so the number reports discovery and only discovery.

The unbranded set starts from buyer language rather than keyword tools. Sales call notes, support tickets, community threads and review-site comparisons show how buyers in a category actually phrase their needs, including the constraints they attach: team size, budget, industry, integrations. Each question should describe a need a real buyer segment has, avoid every vendor name including the brand's own, and be specific enough that the answer functions as a shortlist. Generic questions produce generic answers dominated by the largest players; specific questions are where challengers can realistically win mentions.

The branded set is smaller and follows a pattern: the brand alone, the brand for each major use case, the brand versus each named competitor, the brand's pricing, and the brand's weaknesses. These run on the same cadence as the unbranded set, with the same discipline of repeated runs and stored answers, because branded answers drift too. The output to watch is the answer text itself rather than a mention rate, since the mentions are guaranteed and the value is in what gets said around them. Reviewed monthly, the branded transcripts become a correction backlog: each inaccuracy is a prompt for a published page that puts the correct fact where engines can retrieve it.