Start with the answer, not the score
In Cuebase, open Share of Voice, add the shopper questions and competitor brands you care about, then choose Run probe now. The report records whether a tracked brand was named, whether a returned URL matched its tracked domain, and where its name first appeared among the brands you chose to track.
That is useful evidence, but it is not a sales report. AI answers change, a citation can support a general claim rather than a recommendation, and a link can point to a collection, article, retailer, or product page. When you run an external study, save the answer and URLs with the result before you change a page.
The three signals
| Signal | What you observed | What it does not prove |
|---|---|---|
| Mention | Your brand name appears in the answer text. | That your site was linked, recommended positively, or clicked. |
| Citation | A returned source URL has your tracked root domain. | That the text recommends your brand or a particular product. |
| Product link | For your manual log, a link in the answer or its sources resolves to a specific product page on your domain. | That the product was bought, or that it will appear next time. |
A product link is an extra review field, not a separate Cuebase metric. It may be a citation, but an answer can also contain an inline product link. Keep it separate because it answers a specific question: did the result lead a shopper to one of our product pages?
One saved answer, scored three ways
The following is an annotated extract from Cuebase’s public backpack study captured on 12 October 2026. JanSport was a chosen research baseline, not a Cuebase customer. The full saved answer named JanSport and Herschel. It also returned a JanSport collection URL among several URLs. This is evidence from one ChatGPT API response, not a stable ranking or a shopping recommendation.
Saved-response annotationJanSport SuperBreak Plus Backpack appears in the answer text: count one JanSport mention. The returned jansport.eu/collections/superbreak URL is a brand-owned alternate domain, while the tracked domain was jansport.com: count it as a manual brand-owned alternate-domain source, while Cuebase’s tracked-domain citation remains false. Its path is a collection, so count zero product links. Herschel appears in the answer, but the returned URL is third-party: count a Herschel mention, zero Herschel-domain citations, and zero Herschel product links.
The study reported JanSport as mentioned in 6 of 9 completed answers. The alternate-domain example shows why a manual source log can add context without changing the tracked-domain metric.
Use one row per answer
Do not combine a week of results into a single yes-or-no answer. Cuebase’s CSV export contains the saved metrics; for a manual study, keep the exact prompt, engine, date, answer text, and URLs beside it. Record what each URL actually is. A mention rate is the number of completed answers naming the brand divided by completed answers. A citation rate uses the same denominator for answers with a matching tracked domain.
Download the logGet the AI visibility signal log (CSV). It includes an explicitly illustrative row so you can see how to label a mention, citation, and product link without pretending the sample is your data.
Choose the next task from the evidence
Mention, no citation
Read the wording. If the answer describes a feature that is thin or inaccurate on your site, fix the underlying product facts and page. Do not add claims only because an answer used them.
Citation, no mention
Open the returned page. It may support a background fact rather than put your brand in a recommendation. Check whether the page is current, public, and clear about the product.
Product link
Check the exact product page: title, availability, price, images, material, dimensions, care, and variant details. These are shopper-facing facts worth keeping complete regardless of AI visibility.
Neither signal
Do not conclude that your store has a ranking problem from one run. Add related shopper questions, run them again over time, and compare the returned brands and sources before choosing an improvement.
Keep the limits visible
Share of Voice is a rolling estimate, not a guarantee. In Cuebase, positions rank the first appearance among the brands you chose to track; they are not a universal ranking of every brand in the answer. Citations are matched to the root domains you enter, so a marketplace or publisher is only counted if you deliberately track it.
Reviewed 13 October 2026. Use the report to find pages to inspect and facts to strengthen. Then verify every claim on the actual product record before applying a change. Read the AI share-of-voice guide, 10 shopping questions to track, and the backpacks-under-$100 study. See paid AI visibility tracking.