Some Background on the Metric

When people first start tracking their AI search performance, they usually begin with citation rate — how often their site appears as a source in AI-generated responses. This is a useful starting point, and it gives you a general sense of whether your content is being retrieved at all. But over time, as you get more comfortable with the measurement process, you start to wonder whether your results are actually good or just better than nothing.

That’s where this concept comes in. It’s a way of contextualising your citation data against what else is appearing in the same AI responses — a relative measure rather than an absolute one. If your site is being cited twice out of ten queries, that could mean you’re dominating a space with very few competitors, or it could mean you’re one of twenty sites splitting citations across the same set of queries.

The idea has some parallels to how share of voice works in traditional marketing and media measurement, where the focus is on your proportion of total available attention rather than your absolute numbers.

What the Metric Actually Measures

To understand what it captures, it helps to think about what’s happening when an AI platform like Perplexity answers a question. It pulls sources from across the web and assembles an answer that draws on multiple of them. Each of those sources gets credited in the source panel. On a typical query, there might be seven or eight of them, sometimes more.

If you run ten queries and count up all the source citations across all ten responses, you end up with a pool of perhaps sixty or seventy individual citations. Your site might appear in five of them. The question the metric answers is: what percentage of that total pool is yours?

The calculation itself is fairly simple once you have the data — it’s your citations divided by the total across all domains, turned into a percentage. The harder part is actually collecting the full dataset, because it means recording not just whether your site appeared, but every other site that appeared alongside it.

This is the part that makes the approach more labour-intensive than standard citation rate tracking, and it’s also why most people don’t do it. But the additional data is what makes the metric meaningful, because without knowing the total you have no basis for interpreting your own share.

How to Think About Your Numbers

There’s no universally agreed standard for what a good result looks like, partly because it depends so much on the competitive environment around a particular topic. In a space where AI platforms are regularly drawing from a large and diverse pool of sources, even a well-performing site might only capture a small fraction of the total. In a more specialised area where fewer sources exist, the same site might achieve a much higher proportion without any additional effort.

What tends to be more useful than comparing against an external benchmark is tracking how the number changes over time, and also looking at which specific domains are appearing alongside yours. That second piece of information — who the other sources are — is often what makes the analysis actionable.

If the sites that are appearing more frequently than yours are large, established publishers with significant domain authority, that tells you something different about the situation than if they’re smaller, newer sites covering similar ground. In the first case, the challenge is more about authority and presence. In the second, it’s more likely to be about how your content is structured and how well it matches what AI retrieval systems are looking for.

This distinction matters because the approaches you’d take to address each situation are quite different, and conflating them can lead to putting effort into the wrong things.

The Data Collection Process

Collecting the data for this kind of analysis requires extending your usual measurement routine. Rather than just noting whether your domain appeared in a response, you need to record every domain that appeared — the full source panel for each query. This is more time-consuming, but the information is right there alongside your own results, so it’s a matter of discipline rather than access.

Over a run of ten queries, you might end up with a spreadsheet that has seventy or eighty rows, one for each citation across all responses. From that you can calculate the metric directly, and also start to see which competing domains appear most frequently, which queries have the most concentrated source panels, and which are more diverse.

Some practitioners find it useful to do this kind of extended analysis quarterly rather than monthly, given the additional work involved. Others integrate it into their regular monthly protocol. The right cadence depends on how actively you’re making changes and how quickly you need to detect whether they’re working.

How This Relates to Other GEO Measurements

The metric sits alongside citation rate in your measurement toolkit rather than replacing it. Citation rate is still the primary longitudinal tracking metric — it’s simpler to collect and gives you a consistent signal about your own trajectory over time. The share of voice figure adds the competitive dimension that citation rate lacks.

There’s also a relationship between this kind of competitive measurement and the work of understanding why certain sites perform better than yours on specific queries. Sometimes it points to structural differences in how content is written. Sometimes it reflects domain authority gaps that take longer to address. Sometimes it’s about topic specificity — a site that publishes exclusively about a narrow area may outperform a broader site on queries in that area, even if the broader site has higher overall authority.

Understanding these patterns is part of the longer-term work of improving your position in AI search, which tends to unfold over months rather than weeks and requires combining measurement with structured experimentation.