SEARCH VISIBILITY

How to Separate Branded and Unbranded AI Shopping Questions

Use paired shopper questions and separate denominators so a brand-aware prompt does not inflate your discovery result.

By Cuebase · Published · Updated

Use two series for one buying intent

Branded and unbranded questions can both be useful. They answer different questions. An unbranded question asks whether the assistant surfaces a brand for a need. A branded question asks what happens after a shopper already supplies that brand.

Use the same need, market wording, and constraints. Change only the brand reference. For example:

SeriesPaired promptWhat the result can tell you
Unbranded“What rain jacket would you recommend for cycling to work in the United States?”Whether your fixed brand appears for that shopping need.
Branded“Is Northwind’s rain jacket a good choice for cycling to work in the United States?”How an answer handles Northwind after the shopper supplied its name.

Northwind is a fictional label used only in the example. This guide does not report a live Cuebase result, a merchant result, or a search study.

Keep the denominators separate

In Share of Voice, add each question as a tracked query and add the competitors you want to measure. Choose Run probe now. Cuebase records metrics by saved probe and by tracked brand. Its Export CSV gives long-format rows with run time, query, category, engine, brand, whether the brand was mentioned, whether its tracked domain was cited, and first appearance among the tracked brands.

Calculate each series against its own completed probes. Do not combine the numerator from the branded prompt with the denominator from the unbranded prompt. A failed probe is missing, not a no.

Illustrative seriesCompleted probesBrand mentionsRateCorrect reading
Unbranded311 ÷ 3 = 33%The fictional brand appeared in one of three completed discovery answers.
Branded333 ÷ 3 = 100%The fictional brand appeared after the prompt supplied its name. This is not a discovery rate.

Those numbers are invented solely to show the calculation. They are not a test result and do not prove that adding a brand name causes a mention or citation.

Label the report so nobody merges the signals

  1. Name the series in the query category

    Use labels such as “Discovery — unbranded” and “Brand-aware — branded.” The text of the question remains the evidence; the label makes the dashboard easier to read.

  2. Fix the rest of the setup

    Keep engine availability, competitor set, tracked domains, market wording, and scoring rule constant within a comparison. If one changes, record a new series.

  3. Export the metrics, then retain context

    Use Export CSV for Cuebase’s recorded probe metrics. The normal export does not include raw answer text or a source-by-source citation log. If you inspect a source manually, keep that separate record with its method and date.

  4. Report two lines, not one blended score

    Show the unbranded mention and tracked-domain citation rates beside the branded rates. Write “brand-aware” on the branded line every time it appears in a report.

Use the scorecard

Download the branded/unbranded scorecard (CSV). It includes two clearly labeled fictional rows and one blank row. Start with a few paired prompts, record completed probes, and keep the two denominators separate.

For the metric definitions, read mentions, citations, and product links and your first AI visibility baseline. For the product view, read how Cuebase measures AI share of voice. See paid AI visibility tracking →