Track brand mentions in AI search: build a reproducible baseline

Define the event you are measuring

A brand name in an answer, a linked citation and a recommendation are different events. Record them separately. A citation can point to your site without recommending your product. An answer can recommend the brand without linking to it. Decide which event serves your question before opening a visibility dashboard.

Use a stable set of prompts covering category questions, buying criteria and specific tasks. Record platform, market, date and relevant account context. A result from one prompt on one day is a sample, not a census of everything customers see.

Choose a tracker for the prompts and evidence you need

Semrush documents Prompt Tracking and visibility reports. Its AI Visibility Toolkit page lists $99 per month, 25 tracking prompts and 10 daily CSV exports; adding it to an annual subscription can align billing to that existing term. Consider it when its prompt allowance and reporting scope cover your measurement plan. Verify supported platforms and region settings for the selected report.

Ahrefs documents custom prompts in Brand Radar, alongside its broader prompt dataset. It is a candidate when you want a custom question set and inspection of answer records. Current base-plan requirements, selected add-on price and prompt allocation were not confirmed here. A platform’s large dataset does not prove that it represents your specific audience.

Worked example: correct the denominator and keep citations separate

Completed editorial calculation with a fictional ten-answer sample: three answers name the brand, two of those link to its website, and one explicitly recommends it. Initial report: “We own 30% of the AI market.” That conclusion overstates the evidence and ignores the sampling frame.

Corrected report: “In this fixed ten-answer sample, brand mentions occurred in 3/10 answers, linked citations in 2/10 and explicit recommendations in 1/10.” The rates are 30%, 20% and 10%, respectively. These are calculations on invented observations, not a live measurement of AI Tool Compass. Store the answers so a colleague can inspect how each event was coded.

Keep the prompt panel stable when measuring change

Freeze the baseline prompt list, then rerun it with the same recorded settings. Add new questions in a separate panel so changed coverage is not confused with improved visibility. If the platform, geography or query wording changes, annotate the comparison instead of silently treating it as the same series.

Read the answer text behind a dashboard increase. Check whether the brand is named positively, negatively or only as a comparison. Record the cited page and whether it answers the actual question. A source mention is a lead for content review; it does not by itself identify which change caused the result.

Connect the measurement to a useful action

Use recurring missing topics to improve relevant pages with evidence, clear descriptions and accurate product facts. Monitor your own referral and conversion records separately. Visibility can rise without revenue, and a sampled decline can occur without a meaningful business change. Avoid promising that a tracker itself improves rankings.

Before buying, count the prompt panel, required platforms, export needs and reviewers. An illustrative 20-prompt panel measured across three platforms produces 60 observations per round; the product may bill or sample those differently. Confirm its exact unit rather than calling all 60 “keywords.” Keep the raw records and a dated summary with your action log.

AI search measurement questions