AI Search Brand Mentions:
How to Track Them

Track AI search brand mentions by saving the generated answers in which a brand appears and comparing the same questions over time. Record whether the brand is named, recommended, described accurately, or linked as a source. Add platform citation reports and website referral data to measure those separate outcomes.

A practical starting setup is a spreadsheet of exact questions, a saved answer for each check, access to the relevant webmaster reports, and website analytics. An automated monitor should preserve the same evidence. The key is to make every reported mention traceable to an answer, a date, and a search surface.

What counts as an AI search brand mention?#

Use the following working definitions in the tracking log. They are counting rules for this workflow; individual platforms may use different terminology.

Keep names shown only in source labels separate from names in the answer itself. Also distinguish a recommendation from an incidental reference or a critical description. Save the sentence containing the name so the classification can be reviewed.

Before counting, list accepted company names, product names, abbreviations, and former names. Flag ambiguous matches for manual review. Count each brand once per answer for the metrics below, even when its name appears several times.

How to track brand mentions in AI search#

1. Build a fixed list of questions#

Choose questions relevant to the information the brand should be associated with: category definitions, product comparisons, alternatives, features, availability, pricing, and support. Use existing search queries, site-search records, or documented customer questions where available.

Separate branded questions, which contain the brand’s name, from unbranded questions, which do not. The first group checks how the service describes a named brand. The second checks whether the brand appears without being supplied in the question.

Assign each question an ID, topic, language, and intended search surface. Keep the wording unchanged for recurring checks. Add new questions to a separately labeled group so changes in the question list remain visible in the results.

2. Record the search conditions#

Identify the product and feature being checked, such as ChatGPT with web search, Google AI Overviews, Google AI Mode, or Microsoft Copilot. Record the date, time zone, language, location settings, and model or mode when the interface displays them.

Personalization belongs in the record: OpenAI’s ChatGPT search guidance explains that saved memories may influence rewritten search queries and that approximate location can affect local results. Start recurring checks in a new conversation and keep memory and location settings consistent. Record any context that cannot be controlled.

For Google, log AI Overviews and AI Mode separately when inspecting answers. Mark a search that produces no AI Overview as an availability observation rather than inventing an answer to score.

3. Save the answer and its sources#

Create one row per captured response. Keep the complete answer text and a screenshot or saved capture, then record:

• The exact question and capture date.
• The service, feature, language, and settings.
• Whether the brand appears in the answer text.
• The sentence containing the mention and its role.
• Whether the answer recommends the brand.
• Links to the brand’s website and relevant third-party sources.
• Named competitors, using the same counting rule.
• Specific factual errors and the evidence used to verify them.

In ChatGPT, citations and the Sources view provide links to inspect when available. Open the cited pages before attributing a statement to them; OpenAI warns that search results and citations can be incomplete, outdated, or incorrect.

Review tone separately from factual accuracy. Label the mention favorable, neutral, critical, or mixed only when the surrounding wording supports that description. Keep the quoted sentence alongside the label.

4. Repeat the checks on a consistent schedule#

Choose a weekly or monthly schedule and repeat the same questions under the recorded conditions. Decide in advance how many responses to capture per question and apply that rule across the panel. Preserve each capture instead of overwriting earlier answers.

Compare results within the same service, feature, question group, language, and location. Show missing captures and failed searches separately. When wording, settings, or the monitored service changes, mark the change in the report before interpreting the trend.

Which platform reports help track AI visibility?#

Answer captures show the wording of a mention. Webmaster reports add information about how the website appears on the surfaces those reports cover.

Bing Webmaster Tools: citations and cited pages#

Bing’s AI Performance announcement describes reporting across Microsoft Copilot, AI-generated summaries in Bing, and select partner integrations. It includes total citations, average cited pages, page-level citation counts, and citation trends.

In Bing Webmaster Tools, review the AI Performance report for the selected date range. Save the cited URLs and associated grounding queries. Bing defines these queries as phrases used to retrieve referenced content and describes the displayed query data as a sample of citation activity in its explanation of grounding queries.

Use this report to identify pages receiving citations and topics to inspect in the answer captures. Bing says citation counts do not establish placement, ranking, or a page’s role in an individual answer in its metric definitions. Keep citation totals separate from the count of answers that name or recommend the brand.

Google Search Console: generative AI impressions#

Google’s Generative AI performance report for Search covers impressions from AI Overviews and AI Mode. It provides page, country, date, and device dimensions. The report’s definition of an impression concerns a link to the website being shown in a generative AI feature.

Open the report for the relevant property, select a date range, and inspect the page and country views. Export the data for recurring comparisons. Treat it as a site-link visibility measure; inspect captured answers to determine whether the brand is named in their text.

Google explains in its chart and table guidance that property-level chart totals and page-level table data use different aggregation rules. Avoid adding page rows to reconstruct a property total. The report also excludes Search Labs experiments, as specified in its coverage documentation.

How to calculate mention rate and share of voice#

Calculate metrics from the saved answers, with the question set and counting rule attached. For a spreadsheet, use these definitions:

For sampled share of voice, count each monitored brand once per answer and name the competitor set. Multiple brands can appear in one answer, so the denominator is the total across brands, rather than the number of answers inspected.

Report the raw counts alongside the percentages. Keep branded and unbranded question groups separate, and show capture completion beside each result. A change in the question list or competitor set changes the calculation even when the saved answers remain the same.

These formulas describe the selected answer sample. They do not supply the number of people exposed to the brand. Keep platform-reported citation totals and impressions in their original units instead of combining them with sampled mention rates.

Track visits from AI search in analytics#

Google Analytics’ Traffic acquisition report provides Session source and Session source / medium dimensions, along with sessions, engagement, and key events. Use those dimensions to inspect traffic attributed to identifiable AI services.

In GA4, open Traffic acquisition and switch to Session source / medium. Filter for the AI-service sources present in the property. Review their sessions and key events over the same reporting period, and inspect the pages receiving that traffic. Preserve the source labels rather than assigning every visit to a generic AI category.

For ChatGPT, OpenAI’s publisher FAQ says referral URLs automatically include utm_source=chatgpt.com. Check that the parameter remains intact through redirects and appears in the recorded acquisition data. This identifies tagged inbound traffic; the answer capture supplies the wording and context of a mention.

Google defines (direct) / (none) traffic as traffic without a clear referral source. Leave those sessions labeled as direct unless there is separate evidence of their origin. A rise in direct traffic alone cannot identify visits from AI search.

What to require from an automated tracking tool#

Evaluate an automated monitor against the record that needs to be maintained. Require exact prompt exports, original response captures, citation URLs, timestamps, service and feature labels, language and location settings, repeat-run settings, and a documented brand-matching rule.

Check whether the monitor collects responses through the consumer interface or an API, and label the method in the report. Review a selection of its captures against the reported mentions, recommendations, and sentiment labels. Keep historical prompt lists and disclose any changes to the collection method.

Google states in its official optimization guide that third-party tools do not have access to its internal ranking or AI systems. Assess a monitor through the evidence it exposes and the workflow it supports. Keep its proprietary visibility score separate from the directly counted metrics.

How to use the findings to improve brand visibility#

Prioritize specific findings: repeated incorrect descriptions, unanswered questions, outdated cited pages, and technical access problems. Record the affected question and source URL, make the relevant correction, then repeat the same checks. Treat an increase after an edit as an observation to investigate; the tracking log alone does not establish what caused it.

For Google, the current generative AI optimization guide requires pages to be indexed and eligible for a search snippet, with the site included in generative AI features through Search Console. It also emphasizes useful content and a clear technical structure. Check those requirements before interpreting an absence as a content problem; inclusion is not guaranteed.

For ChatGPT, OpenAI’s crawler documentation distinguishes OAI-SearchBot for search, GPTBot for potential model-training content, and ChatGPT-User for user-triggered actions. Check search access using the relevant crawler controls. Keep bot requests in a crawl-access record: a page request does not contain evidence of the brand appearing in a displayed answer.

The recurring report can stay compact: captured-answer counts, mention and recommendation rates by question group, cited pages, platform impressions or citations, identifiable referral sessions, and the exact statements needing correction. Attach the dates, settings, question list, and answer evidence so the next check can use the same basis.