How do you track brand mentions in AI search?

What does tracking brand mentions in AI search mean?

It means measuring how often AI assistants name your brand when someone asks a question your product answers.

A search engine returns a list of links, and rank tracking tells you where you sit on it. An AI assistant returns a written answer. It either names you or it doesn't, and it may name competitors in the same sentence. Tracking therefore records two things per answer: whether your brand appears, and which other brands appear with it.

The unit you track is a question, such as 'what is the best open source feature flag tool' or 'how do I monitor Postgres query performance'. Keywords are too short to reproduce what a buyer actually types into a chat box.

How do you set up AI search monitoring for your brand?

Write a question list, pick the engines, fix a schedule, and record the same fields every time.

The setup is the same whether you use a spreadsheet or a tool. What matters is that nothing changes between runs except the answers, so the only thing you measure is the engines.

• Step 1: write 10 to 20 questions a buyer would ask before they know your name. Include category questions, problem questions and 'alternative to X' questions for your biggest competitor.
• Step 2: choose the engines your buyers use. For most software buyers that means ChatGPT, Claude, Gemini and Perplexity at minimum.
• Step 3: pick one day a week and run every question in every engine on that day.
• Step 4: for each answer, record the date, engine, question, whether you were named, which competitors were named, and any links the answer cited.
• Step 5: calculate your share: answers that named you divided by all answers. Watch the trend over four or more weeks, not a single reading.

What should you record for each AI answer?

Record enough to compare weeks and to act: the question, the engine, who was named, and the full answer text.

Keep the full answer text. A yes or no tells you that you lost; the text tells you why, for example that the answer described your product as 'enterprise only' or named a competitor for a feature you also have.

How often should you check AI answers?

Weekly is enough for most teams; daily checks mostly measure noise.

AI answers vary from one run to the next even when nothing about your market has changed. Checking daily gives you more readings but not more signal, and it multiplies the manual work. A weekly run on the same day, compared over a month, shows real movement.

MarketHQ runs its sweep weekly, starting each Monday, and shows the most recent week's answers next to a four-week trend line.

How do you integrate AI mention tracking with your existing workflow?

Connect the weekly reading to the work it should change: content, comparison pages and community replies.

A mention report that nobody acts on is a vanity metric. Tie each lost question to one action: publish a clear answer page, fix an outdated comparison, or show up in the threads and lists the engines appear to draw on.

In MarketHQ the weekly AI-visibility reading feeds the gap analysis and action plan, so a question where competitors are named and you are not becomes a task. Paid plans also expose AI visibility and the gap report through the REST API and an MCP server, so your own agents or dashboards can pull the same data.

When is a manual spreadsheet enough?

A spreadsheet works for a handful of questions checked once or twice; it breaks when you want a weekly habit across several engines.

Ten questions across four engines is 40 answers a week to run, read and log. Most teams keep that up for two or three weeks. If you only want a one-off reading, MarketHQ's free LLM mention checker asks ChatGPT and DeepSeek whether they name your domain, with no card required.

When you want the weekly loop, MarketHQ's paid plans track 10 questions on Lite ($29 a month), 60 on Startup ($99) and 120 on Growth ($199), each asked in six engines every week.

What does one week of tracking look like?

Illustrative example: a monitoring tool tracks 10 questions and finds one engine that never names it.

Illustrative example: an observability startup tracks 10 buyer questions. In week one it is named in some ChatGPT and Perplexity answers but in no Claude answers, and the same two competitors appear in almost every answer it misses. The team rewrites its comparison page to state plainly what it does better, answers two open threads where buyers asked the same question, and checks the same 10 questions the following Monday.

What can AI mention tracking not tell you?

It measures a sample of answers, not what every user sees, and it does not cover every AI surface.

Answers change with the account, the location, the conversation so far and the app version, so any tracker gives you a reading, not a census. MarketHQ covers ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek. It does not track Google AI Overviews, Google AI Mode or Microsoft Copilot, so if those matter most to your buyers you will need another tool for them.