Stop Trying to Measure “AI Visibility”

I believe it is practically impossible to reliably measure how often ChatGPT and other large language models recommend a particular brand.

All those online tools that assign a specific percentage of “AI visibility” to a brand are, in my opinion, based on very little. I find them ridiculous.

I also believe that we can avoid trying to measure this metric directly. What really matters is understanding how much traffic LLMs are actually generating for your website. If a significant share of your traffic is coming from LLMs, you should be able to estimate that share from your analytics data.

Measure Traffic, Not AI Visibility

Imagine a page receives 100 visits from real users one month, and 200 the following month. Its traffic has doubled. That is what you should care about.

If all of those visits are coming from LLMs, then you can reasonably conclude that the visibility generated through LLMs has doubled as well.

If only some of the visits come from LLMs, you can estimate the percentage of traffic generated by LLMs by looking at your total traffic and subtracting the traffic you can confidently attribute to other sources. From there, you can arrive at an approximate estimate of the impact of LLMs.

But Not All Unattributed Traffic Comes From LLMs

Of course, there is an important caveat: you cannot simply assume that all traffic you cannot attribute to another source comes from LLMs.

Some of it may be direct traffic. Some may come from apps, messaging platforms, referrers that were not passed correctly, or other sources that are difficult to identify.

So yes, this is an approximate method.

But I still consider it far more reliable than looking at a page and thinking something like: “This page has an H1 here, a link there, so its AI visibility must be X%.”

That is ridiculous.

Measure the Impact of LLMs Instead

It is much more useful to estimate the effect of LLMs on traffic and conversions than to try to estimate how many times an LLM has mentioned or recommended your brand.

Instead of asking:

“How much does ChatGPT recommend me?”

perhaps we should be asking:

“How much traffic and how many conversions are LLMs actually generating for me?”

Approximate Data Is Better Than a Meaningless Metric

In practice, the two questions are trying to get at roughly the same thing: how much exposure and business value are LLMs generating for your brand.

But there is a fundamental difference in how we try to answer them.

The first question tends to lead to a questionable system for estimating an abstract metric such as “AI visibility”, often based on assumptions about how LLMs interpret and recommend a brand.

The second leads to a system that is certainly imperfect and approximate, but is grounded in something that actually matters: real users, real traffic, and real conversions.

And personally, I would much rather have an imperfect estimate based on actual business outcomes than a precise-looking number based on a metric that is almost impossible to measure reliably.