A SaaS marketing lead pulls up ChatGPT, asks it to recommend project management tools for a 50-person team, and their brand does not appear. They check Google. They rank third organically for the exact same query. Two different measurement systems produced two different outcomes, and most teams still only have a dashboard for one of them. Measuring AEO means building the other dashboard, and it starts with knowing which of four numbers you are actually looking at.
Key takeaways
- Mention rate and citation rate answer different questions: whether a model talks about you, and whether it names you as a source.
- Share of voice from the SEO world does not transfer to AI answers, because a brand can dominate the open web and still get near-zero citations.
- Average position inside an AI answer matters because most users read the first one or two brands named and stop there.
What mention rate actually counts
Mention rate is the simplest of the four and the easiest to misread. It measures how often your brand name shows up in an AI model’s response to a relevant query, regardless of whether the model credits a source, links out, or gets the facts right.
Run the same 50 prompts against ChatGPT once a week. If your brand appears in 22 of those 50 responses, your mention rate is 44%. That number tells you the model has some awareness of your brand and category. It does not tell you whether that awareness is accurate, whether it is helping you, or whether the model is even using current information.
Mention rate is a starting point, not a scorecard. A brand can have a high mention rate built entirely on outdated or wrong information, which is worse than a low mention rate built on nothing at all. Use it to establish whether you exist in a model’s frame of reference before you worry about anything more precise.
Citation rate is a different question entirely
Citation rate measures how often a model, when it does mention your brand, actually names you as a source or links to your content. This is where the real distinction between old SEO thinking and AEO thinking shows up.
Perplexity cites its sources inline, so citation rate is straightforward to track there. Other models are less consistent about surfacing sources, which is exactly why citation rate needs to be tracked separately from mention rate rather than assumed to move together. A model can describe your product accurately in a conversational answer and never once point to where that information came from.
A brand can have high share of voice and near-zero share of citation, and those two numbers are measuring completely different things.
This is the concept Answerburst calls share of citation: how often AI models pull a brand into an answer and name it as a source, distinct from share of voice, which is how often a brand gets mentioned across the web overall. A brand with strong PR coverage and heavy social presence can still have a citation rate near zero if none of that coverage is structured in a way models trust enough to attribute.
Citations also compound. Once a model cites a brand for one query, the next citation in a related query becomes more likely, because the model has already treated that source as reliable once. Low citation rate is not just a gap. It is a gap that gets harder to close the longer it sits open.
growth in AI brand presence for NewDay USA, with an average rank of 1.59 on Google AI Overviews, tracked weekly across five engines over six months.See the case study
Share of voice from SEO does not translate cleanly
Teams coming from a search background often try to reuse their share of voice methodology for AEO reporting, and it breaks quickly. Traditional share of voice counts mentions, backlinks, and impressions across the open web, treating volume as the signal that matters.
AI models are not counting volume the same way. They are weighing agreement across independent, high-authority sources against how current and structurally readable that content is. A brand with ten thousand social mentions and thin, unstructured product pages can lose to a competitor with a fraction of the mentions but clean documentation, consistent facts across sources, and recent updates.
The fix is not to abandon share of voice. It is to track it alongside citation rate and treat any gap between the two as a diagnostic. A wide gap, high voice and low citation, usually means the content exists but is not structured or authoritative enough for models to trust it as a source. That is a different fix than a low mention rate problem, and conflating the two wastes effort.
Why average position changes what a mention is worth
Not all mentions carry equal weight. If a model lists six competitors and your brand comes fourth, that is a very different outcome from being named first, even though both count as a mention in a naive count.
Average position tracks where your brand lands across the answers where it appears at all: first named, buried in a longer list, or mentioned only as a caveat after a direct competitor. Users reading AI answers behave the way searchers behave with a list of links. They act on what they see first and rarely scroll deep into an enumerated answer read aloud or displayed in a chat window.
This is where the four metrics stop being independent and start telling a combined story. A brand can have a healthy mention rate, a respectable citation rate, and still be losing the category because its average position sits fourth or fifth every time. Position is often the metric that best predicts whether AI visibility is translating into anything a revenue team would recognize.
Putting the four together without drowning in dashboards
None of these four metrics is diagnostic on its own. Mention rate tells you whether you exist in a model’s answers. Citation rate tells you whether the model treats you as a credible source rather than background noise. Share of voice, tracked against citation rate rather than in place of it, tells you whether your existing web presence is converting into AI trust. Average position tells you whether any of that visibility is worth anything once a user is actually reading.
The models themselves behave differently enough that tracking one and assuming it generalizes is a mistake. ChatGPT pulls from across the open web, Gemini cross-references Search, YouTube, and Scholar, and Claude leans on high-authority publications and documentation. A brand can score well on one engine’s version of citation rate and poorly on another’s, for reasons specific to how each model retrieves and weighs sources.
Build a benchmark across the engines your buyers actually use, run it on a fixed cadence, and read the four numbers as a set. A rising mention rate with a flat citation rate is a structure problem. A strong citation rate with a weak average position is a competitive problem, not a visibility problem, and it needs a different response.
What to do with these numbers this week
Pick 20 to 30 prompts your actual buyers would type or ask, covering comparison questions, “best for X” questions, and direct brand queries. Run them against at least three engines your audience uses, and log four things for each response: whether you appear, whether you are cited as a source, where you rank in the list if there is one, and what the model said about you.
Do that once now as a baseline, then again in four to six weeks. The direction of movement across the four metrics matters more than any single week’s number, and it will tell you whether the gap is a coverage gap, a trust gap, or a competitive one before you spend a quarter fixing the wrong problem.
FAQs
Is a high mention rate enough to say AEO is working?
No. Mention rate only shows that a model talks about your brand. It says nothing about whether the model cites you as a source, positions you favorably against competitors, or is working from accurate, current information. Treat it as a first checkpoint, not a result.
Why would citation rate stay flat while mention rate rises?
This usually means the model has picked up general awareness of your brand from broad web coverage but does not consider your own content structured or authoritative enough to name as a source. It is a signal to invest in structured data, documentation, and content architecture rather than broader PR.
Can share of voice and share of citation move in opposite directions?
Yes, and it happens often. A brand can increase its overall mentions across the web through campaigns or press while its citation rate in AI answers stays flat or drops, because volume and structural trust are measuring different things.
How often should these four metrics be measured?
Weekly tracking catches volatility from model updates, but monthly comparisons are usually enough to judge whether structural changes to your content are working. Match the cadence to how frequently you can act on the results.
