How to measure AEO: metrics that prove AI cites you

The practical move is to measure the answer itself as the unit of visibility. Each metric in the table below isolates one part of that picture, and you can start tracking all of them with a spreadsheet and a recurring block of time before you ever buy a dedicated tool.

VIDEO TRAINING

Get the Growth Playbook.

Learn to plan, budget, and accelerate growth with our exclusive video series. You’ll discover:

• The 5 phases of profitable growth
• 12 core assets all high-growth companies have
• Difference between mediocre marketing and meteoric campaigns

How do you measure how often AI cites you?

Measure citation frequency by running a fixed set of buyer questions through the major AI tools on a regular schedule and logging every time your brand or your content shows up in the response. Pick fifteen to thirty questions your buyers actually ask, word them the way a person would ask an assistant, and keep that list locked so every run is comparable. Then ask each question in ChatGPT, Perplexity, Gemini, and Google's AI Overviews, and record what comes back.

Count two kinds of appearance separately, because they mean different things. A linked citation, where the model names your page as a source, is the strongest signal and the one most likely to pass a click. A mention without a link, where the model states your insight or names your brand inside the prose, still proves the content reached the answer even though it won't show in your referral data. When you log both, you capture the full count, because your analytics on their own only ever see the linked slice.

Tally your appearances as a raw number per run and watch the trend across runs. A jump from showing up in four of twenty questions to eleven of twenty is the clearest evidence your content is landing, and because you held the question set steady, the movement reflects your work rather than a change in what you measured.

How do you measure your share of AI answers?

Measure share of AI answers by recording who gets cited for each question in your set, then expressing your appearances as a percentage of the questions you target. This turns a pile of individual citations into a competitive read, because showing up in twelve of twenty questions means something different depending on whether a rival shows up in three or in eighteen. By measuring share, you give that raw appearance count the context it needs to mean something.

Build it from the same runs you already do for citation frequency, so it costs you no extra testing. For each question, note every source the model cites, including competitors and third-party sites. Over a full run you'll see a pattern: the questions you own, the ones a competitor owns, and the ones where a neutral source like a directory or a review site keeps winning. That breakdown points straight at where your next content effort earns the most ground.

The share view also catches the close calls that raw counts hide. When you and a competitor both appear for the same question, the model usually leans on one of you more heavily or lists one first, and tracking that ordering over time tells you whether you're gaining or losing the position that matters most.

How do you track referral traffic from AI tools?

Track AI referral traffic by filtering your analytics for visits coming from AI assistant domains and watching how those sessions behave once they land. In HubSpot, Google Analytics, or whatever platform you run, look for referrers like perplexity.ai, chatgpt.com, and the source tags Gemini and Copilot attach to outbound clicks. These tools increasingly pass clicks back to the pages they cite, and that traffic is some of the clearest evidence you can get because it confirms a citation turned into a real visit. You can see how that visibility translates into pipeline across our client results.

Set up a saved segment or report for these sources so you're not rebuilding the filter every week. Watch session volume first, then look at what those visitors do once they arrive, because a citation only earns its keep when the traffic it sends sticks around and converts. Referral traffic also serves as a useful cross-check on your manual citation log, because a spike in Perplexity referrals usually lines up with a question you started winning in your prompt tests.

One honest limit is worth naming. Referral traffic only captures the citations that came with a click, so it undercounts your true visibility by leaving out every answer where the model used your insight without sending anyone over. That's exactly why you run the manual prompt tests alongside it, since the analytics by themselves can only ever tell you part of the story.

How do you use prompt tests to track AEO over time?

Use prompt tests by re-running your locked question set on a fixed cadence and comparing each run against the last, so you can see whether your presence is holding, climbing, or slipping. The discipline that makes this work is consistency: keep the exact wording of every prompt, run them in the same tools, and space them on a regular schedule like monthly or every few weeks. Any change to the prompts breaks the comparison, since you'd no longer know whether the movement came from your content or from a reworded question.

Record three fields for each prompt on each run: whether you appeared, where you appeared relative to other sources, and how the model framed your content. Stored run over run, these fields turn into a time series you can actually read, and the pattern usually shows up well before any single run looks dramatic. A question you've climbed in for three runs straight is a page doing its job, while one you've slipped in is a signal to revisit the answer or its structure.

For a deeper read on the testing side specifically, the same prompt-set discipline that powers featured snippet work applies here, and we've written about how to test for a Google featured snippet using a comparable repeatable approach.

How do you measure whether AI framed your content accurately?

Measure framing accuracy by reading every answer that cites or mentions you and grading whether the model represented your point correctly, partially, or wrongly. A citation that distorts your insight can actively work against you, because it puts a version of your expertise in front of a buyer that you didn't write and wouldn't stand behind. Accuracy of framing is the metric that catches this, and it's one a tool can't grade for you yet, so it stays a manual read.

Grade each cited answer on a simple three-point scale. An accurate framing states your point the way you'd state it. A partial framing gets the gist but drops a qualifier or a number that changes the meaning. A wrong framing attributes something to you that your content doesn't actually say. Logging these alongside your citation count tells you how often you appear and, just as usefully, whether appearing is actually helping you.

When the framing comes back partial or wrong, the fix almost always lives on your page, since the page is the part you can actually edit and the model will keep paraphrasing whatever it finds there. An answer that's easy to misread usually has the key point buried, hedged, or split across paragraphs, and tightening it into a single clean, standalone statement gives the model less room to paraphrase you into something you didn't mean.

How do you set an AEO baseline?

Set your baseline by running every metric above once, before you change a single page, and saving the results as your starting point. The baseline is what makes everything else measurable, because a citation count means little on its own and a great deal when you can see it doubled from where you began. Run the full question set, log citation frequency, calculate share of answers, capture current AI referral traffic, and grade the framing on the answers you already appear in.

Date the baseline and store it where you'll find it again, then schedule the re-run. Most teams we work with settle on a monthly cadence, which is frequent enough to catch movement and spaced enough that the engines have time to re-crawl and re-rank between checks. Each subsequent run gets compared against the baseline and against the run before it, so you're reading both the long arc and the recent change at once.

The caveat to hold onto is that AEO measurement is younger than SEO measurement, and the tooling is still maturing, so you read these numbers as directional signals that show you the shape of your progress even when the precise figures stay soft. Citation counts shift run to run partly because the models themselves update on their own schedule, which means a single down month isn't necessarily a problem with your content. What you can trust is the trend across several runs, and the pages that climb in it are reliably the ones that answer a real question cleanly and back it with specifics a model can verify.

Schema markup recommendations

For a measurement guide structured as connected question-and-answer sections, the right structured data gives answer engines a cleaner read on how your content is organized, and we recommend implementing:

• FAQPage schema for the question-based sections (what AEO measurement tracks, how to measure citation frequency, how to track referral traffic, how to set a baseline), since these map directly to how people query AI assistants
• HowTo schema for the baseline-and-cadence workflow, with each step from locking the question set through scheduling the re-run laid out so a model can read the sequence cleanly
• Article schema with author, datePublished, and dateModified fields to signal authorship and recency, both of which answer engines weigh when deciding whom to cite
• Organization schema linking to the Lean Labs brand entity to reinforce topical authority across your AEO content cluster