Without metrics, AEO is an opinion contest. These four turn AI answers into something you can present to a board.
1. Mention rate
Share of your prompt set where the AI names your brand. It is the baseline: 15 prompts, 6 mentions, 40%.
Realistic target: +10 points per quarter on a stable prompt set.
2. Average position inside the answer
Opening the answer is not the same as being seventh in a list. Record where your brand appears in the order the AI names it and average it. Going from 4.1 to 2.3 is worth more than two extra mentions.
3. Share of voice vs competitors
Count every brand mentioned across your answer set and compute your percentage. This is the metric that communicates the problem upward: "in our category, AI names competitors three times for every time it names us".
4. Source share
Which domains the AI cites, and how many of them feature you. This is the actionable one: it tells you exactly where to work next month.
Rules that keep the numbers trustworthy
- Frozen prompt set. Change the prompts and you break the series. Add new ones as a separate block.
- Same cadence. Weekly or monthly, always on the same day.
- Store the literal answer. Numbers without the verbatim quote cannot be audited.
- Multiple assistants. Results differ noticeably between models; track at least two.
The report people actually read
A dashboard with those four metrics plus one plain-language paragraph explaining what changed and three concrete actions. Nothing more. Twenty-chart panels do not change decisions; one clear sentence does.