Track brand mentions in AI search

What you are actually measuring

"AI visibility" usually bundles three different observations. Keep them apart in your notes:

• Mention: the answer names your brand.
• Citation: the answer links to one of your pages as a source.
• Position: where the mention appears, for example first in a list of recommendations.

A brand can be mentioned without being cited, and a page can be cited for a fact without the brand being recommended. AI share of voice is the share of answers to your prompt set that mention you, usually set beside the same figure for each competitor.

This guide covers the method shared by every engine. Each engine also has its own settings, source displays and API fields, covered in separate guides for ChatGPT, Perplexity and Gemini.

Step 1: Write the questions your buyers ask

The prompt set decides what your numbers mean, so write it before you collect anything.

• Discovery prompts that do not name you: "What are the best accounting tools for a two-person agency?"
• Comparison prompts that name you and a competitor: "How does [your brand] compare with [competitor] for invoicing?"
• Brand prompts that check facts: "What does [your brand] do and who is it for?"

Keep the wording fixed between runs, note the language and country, and list the competitors you expect to see. Ten to forty prompts that reflect real buying situations are more useful than hundreds of variations on one phrase. Our industry playbooks include example prompts by sector.

Step 2: Choose how to collect answers

By hand in the apps. Free and closest to what a person sees, but slow, and your own history and location can influence the answer. Use a clean session and write down the date, app and any settings.

Through the official APIs with web search turned on. Repeatable and easy to log. Each provider documents a web-grounded mode that returns sources:

API answers are not guaranteed to match the consumer apps. They are a consistent instrument, which is what trend tracking needs. If you show API answers to anyone else, follow each provider's display terms: OpenAI and Anthropic both require visible citations when web results are shown to end users.

Step 3: Sample repeatedly and keep the raw answers

For every run, store:

• the prompt, engine, model or setting, date and country;
• the full answer text;
• every cited URL;
• whether your brand and each competitor are mentioned, and in what position.

Report a mention rate as "mentioned in 7 of 24 answers", not "29%". The raw count shows how much evidence sits behind the figure, and the stored answers let someone check a surprising result later. Weekly runs are a sensible default: often enough to see the effect of a site change, not so often that cost and noise dominate.

Step 4: Read the numbers with care

• Answers vary. The same prompt can produce different recommendations minutes apart. Look for changes that persist across several runs.
• Engines differ. A brand can be common in one engine's answers and absent from another's, because each searches and ranks sources differently.
• Prompts drift. Changing the wording changes the measurement. Version your prompt set and start a new series when you edit it.
• Correlation is not cause. A rise after a content change is encouraging, not proof. Note what else changed, including competitor launches and news.

Step 5: Use the free reports you already have

Two search platforms report part of this directly:

• Google Search Console counts clicks and impressions from AI Overviews and AI Mode in the standard Performance report under the Web search type. Since June 2026 it also has a generative AI performance report, available to all sites since 2026-08-31, that shows impressions in AI Overviews, AI Mode and Discover by page, country, device and date. It counts impressions only and does not cover the Gemini app.
• Bing Webmaster Tools has an AI Performance report, in public preview since February 2026, showing when your pages are cited in Microsoft Copilot and Bing's AI-generated answers.

Neither shows ChatGPT, Claude, Perplexity or the Gemini app, which is why prompt sampling is still needed.

When you are not mentioned

Check access before rewriting content. If an engine's search crawler cannot reach the page, no amount of copywriting helps. Run the GEO audit checklist or the free AI visibility checker on the pages you expect to be cited, confirm the crawlers in our AI crawler list are allowed, then look at the sources the engines cite instead of you. Those pages show what kind of evidence the engine found useful for that question.

CiteVisor's paid plans automate this method: a fixed prompt set sampled weekly through the four APIs above, with answers, citations and week-over-week changes kept for each run.

Free spreadsheet template (fields)

Use one row per answer so you can return to the evidence behind a percentage. These fields work in an ordinary spreadsheet; the example is illustrative, not a measured result.

For an illustrative completed batch of 20 answers, six brand mentions mean 6/20 mention rate. If three answers link to your pages, citation rate is 3/20. Count each brand once per answer; competitor rates can overlap and need not add to 100%. Compare identical prompt sets before interpreting a change.

Related: how to get cited (not just tracked)

Once you find a gap, use how to get cited in AI search for the shared baseline and how to get cited by ChatGPT search for engine-specific checks. If you are choosing a monitoring platform, compare Ahrefs Brand Radar alternatives by prompt control and retained evidence.

Related guides

• Track brand mentions in ChatGPT.
• Track mentions in Perplexity and get cited by Perplexity.
• Track mentions in Gemini.
• Check access with GPTBot vs OAI-SearchBot and the GEO audit checklist, then run the free AI visibility checker.

Sources

• OpenAI: Web search tool
• Google AI for Developers: Grounding with Google Search
• Anthropic: Web search tool
• Perplexity: Web search in the Agent API
• Google Search Central: AI features and your website
• Google Search Central Blog: Search generative AI performance reports
• Microsoft Bing: Introducing AI Performance in Bing Webmaster Tools

Reviewed on 2026-10-21; the Bing Webmaster Tools details were last checked on 2026-10-09. API features and display terms change; check each provider's current documentation before building a tracker.