How to Measure Your Brand's AI Citation Share Across LLMs

The question every CMO is now being asked

A Series B CMO put it plainly: organic search traffic is flattening, buyers are researching software inside ChatGPT and Claude, and the board wants to know whether the brand shows up there. The honest first answer for most companies is that they have no idea, because they have never measured it.

That gap is the problem. You cannot manage what you do not measure, and "are we visible in AI answers" has felt unmeasurable. It is not. AI citation share is a concrete number, and the method to produce it is closer to a survey than to a guess. Getting your brand into those answers is a separate discipline, covered in how to get your brand cited by ChatGPT, Claude, and Perplexity. This article is about the measurement underneath it: the metric you report, defend, and trend over time.

What AI citation share actually is

AI citation share is the share of a defined prompt set where an answer engine names your brand. If you test 60 buyer-relevant prompts across four models and your brand appears in 9 of them, your citation share is 15 percent. Run the same 60 prompts next month and the number moves, and now you have a trend instead of an anecdote.

The metric only means something if three things are fixed:

• The prompt set is locked and versioned, so month-over-month numbers compare like with like.
• The models are the same each run, so you are not comparing ChatGPT in March to Gemini in April.
• The scoring rule is written down, so two people measuring the same answer record the same result.

Without those, you have noise that looks like data. With them, you have the AI-era version of share of voice.

How to measure it in five steps

1. Build the prompt set from real buyer questions. Start with the questions buyers actually type, not keywords. The strongest source is the language your prospects use in sales calls and the questions an AI-visibility audit already tested. Aim for 40 to 80 prompts that span your category, your use cases, and the comparison questions ("best tool for X", "alternatives to Y").

2. Decide what counts as a citation. Write a one-line rule. A reasonable default: the brand is named in the answer body, not only in a link list. Then add a position tag for each hit, because where you appear matters.

3. Run the set across every engine that matters. At minimum ChatGPT, Claude, Gemini, and Perplexity. Run each prompt fresh, record the answer, and tag whether your brand appears and in what position. Structuring your content so engines can lift it cleanly is its own workstream, and it is what the AI Feed Engine is built to do.

4. Benchmark named competitors on the same prompts. Internal trend lines without competitor context produce reassuring graphs and bad strategy. Your citation share can double while a competitor's triples. Pick three to five named rivals at the start and score them on every run, then add a competitor-delta column to the report.

5. Put it on a calendar. Stakeholders treat irregular measurement as anecdote and scheduled measurement as fact. Monthly is enough for most categories. The cadence is the credibility, not any single run.

Turning it into a board metric

A board does not want a screenshot of ChatGPT. It wants a number with a denominator, a trend, and a competitor line. Three figures carry the story:

• Citation share this period, against the same number last period.
• Hero-citation share, because being the recommended answer is the goal, not being a footnote.
• The gap to your closest tracked competitor, in points.

That is a slide finance can read. It reframes AI visibility from a vibe into a managed line item, the same way keyword rankings did for SEO a decade ago. When OnlyAEO ran this model for a B2B client, the measurement is what unlocked the budget to fix the gaps, the pattern documented in the FastTrackr AI case study.

The mistakes that make the number lie

The fastest way to lose board trust is a metric that cannot be reproduced. Three failure modes do most of the damage:

• Counting every mention as equal, so a trailing link inflates the number that should reflect hero citations.
• Moving the prompt set between runs, which turns a measurement into a coincidence.
• Measuring yourself but not competitors, so a relative loss reads as an absolute win.

Fix the methodology before you spend a dollar on content. A defensible empty program beats an impressive one you cannot reproduce. If you want a free starting point on the structured-data side, the llms.txt generator produces a file that helps AI crawlers find and parse your key pages.

Where to take it from here

Measurement is step one. Once you can see your citation share and the prompts where you lose, the work becomes closing those gaps with content built to be the answer, which is the core of how OnlyAEO works. The teams that win treat AI visibility like any other managed channel: a locked metric, a monthly cadence, a competitor benchmark, and a backlog of gaps ranked by buyer value.