Best Ways to Track Brand Mentions in AI Search

AI answer observation worksheet

Define the event before choosing a dashboard.

Illustrative observation: ask a platform what is included in a studio’s website package. Save the exact question, date, answer and cited URL, then compare the answer with the approved package page. A mention without a link, a linked source and an identifiable visit are three different records. This is a measurement example, not an observed result.

Write the definition beside each metric and keep the source system's wording. If a platform changes its reporting, record the date and avoid comparing unlike periods without a note. This simple discipline prevents a rising impression count from being presented as rising revenue.

Use a controlled prompt sample for questions platforms do not report.

Build a small set of prompts based on real buyer decisions: branded fact checks, category discovery, problem questions, comparisons, pricing or scope questions, and location-sensitive queries where the business has a legitimate market. Keep the wording stable enough to compare snapshots.

For each observation, record the exact prompt, answer, cited pages, platform or experience, date, market, device, and whether the session was signed in. Review factual accuracy as well as appearance. A wrong price or an obsolete service description is more urgent than a missing generic mention.

Start with platform-native visibility data when it exists.

Google's Generative AI performance report in Search Console provides a dedicated view of URL visibility within generative AI features on Search and Discover. The available dimensions include impressions, pages, countries, devices for Search, and dates. Use that data according to Google's definitions and compare the pages that appear with the pages the business considers important.

Native reports have better access to their own platform data than external observation tools, but they still describe only that platform. Keep separate reports for separate systems and avoid calling one dataset total AI search share.

Track identifiable AI referral traffic without overstating coverage.

Segment referrals and campaign parameters associated with AI products, then review landing pages, engaged sessions, useful actions, and enquiries. OpenAI says links from ChatGPT search results include utm_source=chatgpt.com, which can support a specific analytics segment when the parameter reaches the site.

Referral data is valuable because it connects visibility with on-site behavior. It is still incomplete. A person may read an answer without clicking, copy a URL, change devices, or arrive later through another channel. Report measurable AI referrals, not every visit influenced by AI.

Connect visibility to accurate pages and qualified enquiries.

When a cited page receives visits, inspect whether it answers the original question and offers a relevant next step. An informational guide might lead to a service page, a comparison, or a contact form. The path should be useful rather than forcing every reader into the same CTA.

For enquiries, preserve the landing page and available source data, then let a human judge relevance. A form submission does not become qualified simply because the visit came from an AI product. Record the service need, market, project fit, and next sales step using the same criteria applied to other channels.

Review monthly, then investigate material changes.

A monthly review is usually enough for a small business site unless a major launch, crawler change, platform release, or factual error requires faster attention. Compare the same prompt set, platform report, referral segment, landing pages, and enquiry definitions. Save the baseline before changing content.

When visibility changes, look for a corresponding site or platform event. Check indexing, crawler access, page updates, internal links, factual consistency, cited URLs, and competitor or source changes. Do not rewrite several pages because one answer varied once. Use repeated evidence to decide what needs correction.