Is your brand showing up in AI answers? Five steps to build an AI visibility dashboard

Last week, a client came in for a meeting. The marketing director turned his laptop around. 'Our product ranks first on Google. But when I asked ChatGPT what are the best XX solutions available, it listed five companies. We weren't on it. So what should I worry about?'

It's the question I hear most often now. Most companies can't answer it, not because it's hard, but because they're not measuring it. Google rankings have GSC and Ahrefs as clear measurement tools. But how often does your brand show up in AI answers? Ninety percent of marketing teams rely on occasional spot checks and gut feelings. That's not measurement. That's intuition.

This piece explains a repeatable weekly tracking process. You'll turn 'am I showing up in AI answers' into a measurable metric you can monitor. You can start in five steps.

Step one: Define your question set

Don't ask 'tell me about our company'. That just gives the model your name to cite. You need questions that actual buyers ask during their decision process, without mentioning your brand in the query itself.

We typically help clients build 15 to 30 questions in three categories. Category questions ask 'which enterprise RAG implementation consultants operate in Taiwan?' Scenario questions ask 'where should manufacturers look when deploying AI customer service?' Comparison questions ask 'which solution works better for SMBs, A or B?' Once you define your question set, run it the same way every week for consistency. Any changes to the questions render your numbers incomparable.

Step two: Design your repeatable testing framework

Repeatability is essential. Use the same prompt for every question. For instance: 'Recommend 3 to 5 [category] vendors and briefly describe each one's key differences.' Test across models: ChatGPT, Gemini, Perplexity, Claude. If you're tracking the Chinese market, add Doubao and Ernie. Run the same questions on the same day with the same wording each week.

We've encountered two issues. One is memory and personalization. Test in incognito or logged-out mode, or the model will use your chat history and skew the results. The other is LLM instability. LLMs produce different outputs on the same input, so one run per question won't be reliable. Run each question three times and record whether you appear in the majority of responses.

Step three: Define your tracking fields

The real measurement work happens here. Record these fields for each question run:

Calculate your mention rate as (questions where you appear) divided by (total questions tested). This is your key metric. From there, you can calculate mention rates per platform and average position.

Step four: Build your weekly dashboard

A Google Sheet works fine to start. Pick one day each week to run the full test suite, record that week's mention rate, average position, and competitor appearances, then chart it as a line graph. Watch the trend, not individual weeks. If you update your FAQ or publish structured content on a Tuesday, check whether the line goes up three weeks later. When you're ready, you can automate this with an API.

Step five: Connect your numbers to content action

Measurement is only half the work. The numbers only matter if you act on them. When you lose to competitors on certain questions, that gap usually indicates missing comparative content. When the model mentions you but provides no link, it means the model knows about you but lacks a source to cite. That's where structured data and third-party reviews help. Make one content change per week based on your numbers. In two months, the trend line will show which changes worked.

This is how we work with finance and manufacturing clients on their AI visibility initiatives. We don't just deliver a report. We embed the tracking fields into their weekly meetings, so mention rate becomes a regular KPI. The dashboard itself doesn't matter much. What matters is someone looking at it each week and adjusting content accordingly.