How to Measure AI Brand Visibility (And What the Numbers Actually Mean)

ChatGPT, Claude, and Perplexity do not send referral headers when a user clicks a link from an AI response. Even when they do, traffic often arrives without UTM parameters. The result: AI-sourced visitors look identical to someone who typed your URL directly.

This is not a niche edge case. A growing share of B2B software research starts in an AI assistant rather than a search engine. Buyers ask "what's the best [category] tool" or "how do teams handle [problem]" and act on what the model recommends. If you are not in those answers, you are not in the consideration set, and you will never see it in your GA4.

Some teams try to infer AI traffic by looking for unexplained direct traffic spikes, or by filtering for sessions from Perplexity (which does pass some referral data). These are useful supporting signals. They are not a measurement strategy. The only way to know if you are being mentioned is to ask the models directly.

Prompt coverage is the foundation. Of all the prompts relevant to your product category, how many include a mention of your brand? If there are 50 prompts that match how your buyers search, and your brand shows up in 10 of them, your prompt coverage is 20%. If coverage is near zero, nothing else matters yet.

Share of voice tells you how you compare. Out of the AI responses to your target prompts, what percentage mention you versus competitors? If your brand appears in 15 answers and the category leader appears in 35, you have roughly 30% share of voice in that set. Buyers do not choose in a vacuum. Share of voice tells you whether you are in the room when the comparison happens.

Position is the qualitative layer. When you are mentioned, where do you appear? First in the list, or fourth? Framed as the recommended option, or hedged with "some teams also use"? This is harder to quantify but easy to spot when you read the responses directly. A fourth-place mention in passing is not the same as being named first.

The manual approach works well for a first pass. Pick 15-20 prompts that reflect how your buyers actually search: "best [category] for [use case]," "[problem] software comparison," "[alternative to competitor]." Run them in ChatGPT and Perplexity. Screenshot each response. Record in a spreadsheet: did your brand appear, what position, what framing.

This takes a few hours and gives you a real baseline. The limit is scale. You cannot manually track hundreds of prompts, and models update their knowledge over time. What is true today may not be true in six weeks.

The automated approach, using tools like SEOforGPT, runs prompts at scale and tracks responses over time. The value is not the initial snapshot. It is the trend. You want to know whether your content investments are actually shifting your visibility, and for that you need longitudinal data across the same prompt set, tested consistently month over month.

Most B2B SaaS brands start at 0-20% prompt coverage when they first measure. This is not a failure. It reflects the reality that AI models draw on a narrow set of authoritative sources, and most brands have not yet built the content footprint and citation profile that earns inclusion.

40% or higher prompt coverage is genuinely strong for a competitive category. If you are above that, your brand has real authority in the AI knowledge graph.

The goal is not 100%. Many prompts in a category are irrelevant to your actual buyers. A project management tool for enterprise teams does not need to appear when someone asks about "free task apps for students." Coverage of the 30-50 prompts that match your ICP matters more than blanket presence across every tangentially related query.

Pick a fixed set of prompts: 20-30 that represent how your actual buyers search. Run them once a month, in the same models, at roughly the same time. Monthly cadence works because model updates do not happen daily, and trends need a few cycles to confirm.

What you are watching for is directional movement. Coverage going from 15% to 25% over three months is a signal your content work is landing. Coverage flat or declining while a competitor rises means they are building authority faster in the spaces that matter.

A single snapshot tells you almost nothing. The slope is what matters. Set the baseline, run the same prompts, look for movement.

For the practical side of building out your prompt set and tracking process, see our guide to measuring AI visibility and LLM visibility basics.