How to measure AI visibility without fooling yourself
Most claims about “AI visibility” can’t be checked, because there’s no shared, repeatable way most people measure it. Here’s the method I use: a documented prompt panel, what to record each time, and — just as importantly — what it can’t prove.
Why this is harder to measure than search rankings
A search ranking is a stable, checkable number. An AI answer isn’t: the same question can return a different answer depending on the platform, the model version, the date, and even small changes in phrasing. Anyone offering a single “AI visibility score” without naming the platform, model and date behind it is offering an opinion dressed as a metric.
The method: a repeatable prompt panel
- Write down 10 to 20 real buyer questions. Not generic terms — the actual questions your buyers ask when researching, including comparisons (“X vs Y”) and shortlist questions (“who does this in Y”).
- Run the same panel across the main assistants. ChatGPT, Perplexity, Gemini and Google’s AI Overviews cover most of the traffic that matters for a B2B buyer.
- Record five things for each result: the platform, model version and date; whether you’re mentioned; whether you’re cited as a source; whether the description is accurate; and who appears instead.
- Repeat on a fixed schedule. Monthly is usually enough to see a genuine trend without chasing daily noise.
This is the same panel I run as part of an AI visibility audit — the audit is this method applied once, with a baseline and a 90-day plan; ongoing tracking is the same panel repeated.
What to track alongside it
- Mentions — does the assistant name you at all?
- Citations — does it link to or reference your site as a source?
- Accuracy — is what it says about you actually correct?
- AI referral traffic — add a segment in Google Analytics for traffic from chatgpt.com, perplexity.ai, gemini.google.com and similar, so you can see the visits that do click through.
What this can’t prove
Be honest about the limits, because overclaiming here is where most “AI visibility” offers fall down:
- A single audit is a snapshot, not a trend. One run tells you where you stand today, not whether you’re improving.
- AI referral traffic will always undercount real visibility, because many answers satisfy the question without a click.
- No one can guarantee inclusion. The tools and models change on their own schedule. What you can do is remove the things that make you hard to find, verify and quote, and measure the effect honestly over time.
- Small sample sizes are noisy. A panel of 10 prompts can shift for reasons that have nothing to do with your content. Track the trend across months, not the week-to-week wobble.
Where to start
If you’ve never run this kind of check, the fastest way to get a genuine baseline is an AI visibility audit — the same method above, done properly, with a prioritised plan attached. More about AI visibility and GEO →
FAQs
How do you measure AI visibility?
With a fixed panel of prompts that reflect how your buyers actually ask about your services, run across the main AI assistants on a set schedule. For each one, record whether you’re mentioned, cited and described correctly, and who appears instead. Repeating the same panel monthly shows the trend.
Can AI visibility be guaranteed?
No, and be wary of anyone who says otherwise. AI answers vary by model, version, date and even phrasing, and the tools change often. What a credible audit can do is fix the things that make you hard to find, verify and quote, and measure the effect over time.
What should I record in a prompt panel?
For each prompt: the platform, model version and date; whether you were mentioned; whether you were cited as a source; whether the description was accurate; and who appeared instead of or alongside you. Without the platform, version and date, a result can’t be compared to anything later.
Does AI referral traffic prove AI visibility is working?
It’s one useful signal, not proof on its own. Many AI answers satisfy the question without a click at all, so referral traffic will always undercount real visibility. Treat it as a lower bound, alongside the prompt-panel results, not a complete picture.
The Point, once a month
A short note on what’s worth doing in marketing operations, SEO and AI visibility, and what to stop.
Let’s talk
A 30-minute discovery call is free and there’s no pitch. Bring the situation as it is. If I’m not the right fit, I’ll say so, and suggest who might be.