How to Record Your First AI Visibility Baseline

A baseline is a dated starting record

AI answers change. A brand can appear in one answer and disappear from the next. That does not make measurement pointless. It means you need a fixed question set, a clear denominator, and a record of what happened before you alter your catalog.

For each successful answer, capture three separate observations: whether your brand was named, whether your configured brand domain appeared in the answer’s citations, and whether a URL for a specific product appeared. Those are different events. A brand can be named without a citation. A citation can point to a category page. A product URL can appear without proving that the assistant recommended the product.

Get the blank baseline log (CSV). See the filled illustrative example (CSV). The filled file uses invented observations only; it is a format example, not a Cuebase study or a merchant result.

Choose questions before you run anything

Start with six to ten questions a shopper could ask without knowing your brand. Keep the wording intact for the next run. A useful first set usually includes a broad discovery question, a use-case question, a budget question, and a comparison question.

• Discovery: “linen sheets for hot sleepers”
• Budget: “linen sheets under $200”
• Use case: “linen sheets for summer humidity”
• Comparison: “linen vs cotton sheets for hot sleepers”

Do not put your brand name in these first questions. The point is to observe whether you are recommended for the shopping need, not whether an assistant can repeat a name supplied in the prompt. For help building the set, use the 10 shopping-question template.

Run the fixed set in Cuebase

1. Add queries and competitors

   Open Share of Voice. Add each question as a tracked query and add the competitors you want to compare. Cuebase stores the question text and category, so use a category such as discovery, budget, use case, or comparison.

2. Check your brand settings

   Use the brand name and brand domain that identify your store. Cuebase detects a citation by looking for the configured domain in the citation URLs it extracts. That is a domain-citation measure, not a claim that every link reaches a product page.

3. Run the probe and save the date

   Select Run probe now. Cuebase runs one probe for each active query and available engine within your plan, then records successful results. Save the run date, the queries, the competitors, and the engines used. If you change one of those later, you have started a different comparison set.

The report separates mention rate, average position, citation rate, and share of voice. Its CSV export gives one row per brand per saved probe. Use it as your baseline record. If you need to document a specific product URL, inspect the answer and citation URLs separately and write that manual observation in your log. Cuebase does not present “product link” as a dashboard metric.

Use denominators that match the claim

Count failed probes, but do not treat them as a “no.” They are missing observations. If one of six planned probes fails, report five successful probes and one failure. Calculate rates from the five successful probes.

These numbers are computed from the downloaded illustrative log. They do not predict traffic, sales, or a future answer. Their value is the comparison: rerun the same set after a meaningful catalog change and compare like with like. For the broader definitions and limits, read how Cuebase measures AI share of voice.

Keep the first record small and honest

A first baseline is not a verdict on your brand. It is a record of a particular set of questions, engines, settings, and dates. Save the raw export and your manual log together. Do not merge a failed answer into the zero-mention count. Do not call a citation a product click. Do not compare a new question set with last month’s old one.

Then choose one change you can verify. You might prioritize products with the first-10 worksheet before improving a thin product description. After you make that change, run the exact same baseline set again.

If you want an example of how Cuebase reports repeated probes, see the clearly dated backpack search study. It is a one-day study, not a benchmark for your store.

See paid AI visibility tracking →

Source and method notes

1. Cuebase Share of Voice records one successful result per probe, where a probe is an active query sent to an available engine. The product’s report computes mention and tracked-domain citation rates over saved successful results; failures are counted separately by the runner.
2. The illustrative CSV contains invented rows solely to show the log fields and calculations. It makes no claim about an assistant, merchant, competitor, or shopper outcome.