The AI visibility KPIs worth reporting to a CMO

A CMO does not want an AI visibility dashboard. They want to know whether the company is being recommended, whether that's improving, who's beating them, and what it costs to fix. Five metrics answer that. Most AI visibility reporting fails not because the data is bad but because it hands an exec twelve numbers and no decision.

The five KPIs that survive an exec meeting

Mention rate

What it is: the percentage of your tracked prompts where your brand appears at all, averaged across runs.

Why it earns a slot: it's the simplest possible read on presence. If it's 12%, you are effectively absent from AI answers in your category, and nothing else on the slide matters until that changes.

How to compute it: appearances / total prompt-runs. If your 25 prompts run three times each on four engines, that's 300 prompt-runs. Appear in 96 of them and your mention rate is 32%.

How it gets abused: reported as a single blended number across engines, hiding the fact that you're at 60% in ChatGPT and 4% in Perplexity. Report it per engine.

Share of Voice against named competitors

What it is: your appearances as a fraction of all brand appearances on the same prompt set.

Why it's the headline metric: it's the only one that answers "are we winning." Mention rate can climb 13 points while Share of Voice moves 1.7, because everyone else got more visible too. That divergence is invisible unless you track both, and it's the difference between a real gain and a rising tide. The full arithmetic, and the four ways people inflate the number, is in the Share of Voice breakdown.

How to present it: a line chart, you and your top four competitors, monthly, over at least six months. This is the chart. If you make one slide, make this one.

Average citation position

What it is: where your brand lands in the answer when it does appear. First brand named, third, seventh.

Why it matters: appearing seventh in a list of nine is not appearing. Buyers read the first two or three names and stop, exactly as they do with search results. Two brands at identical Share of Voice can have completely different commercial outcomes if one is consistently named first.

How to present it: average rank, with the percentage of appearances that are in the top three. "We appear in 40% of prompts, and 62% of those appearances are top-three" is a sentence a CMO understands instantly.

Sentiment of mentions

What it is: the split of your appearances into positive, neutral and negative framing.

Why it matters more than people expect: an AI engine saying "Acme is popular but users report frequent outages" is a negative mention delivered in the neutral register of an authoritative assistant. That does more damage than a bad review, because it doesn't read like an opinion. It reads like a fact the machine looked up.

How to present it: percentage of mentions that are non-positive, plus the actual sentence when it's bad. Show the sentence. Nothing focuses a leadership team like reading the exact words the model uses about them.

Prompt coverage by buyer stage

What it is: your mention rate broken out by prompt archetype: category, comparison, alternatives, problem-led, jobs-to-be-done.

Why it's on the list: it turns a metric into a roadmap. It's the difference between "we're at 32%" and "we're at 55% on category questions and 8% on alternatives questions, which is where deals are decided." The second version tells the CMO what to fund.

How to present it: a five-row table with your rate and your best competitor's rate per archetype. The rows where you're behind are your budget request. Building the archetype split is a function of how you constructed your prompt set, so if the set isn't right, this KPI can't be either.

The KPIs to leave out

The referral-sessions one deserves a moment, because it's the trap most likely to kill an AI visibility program. Pew Research found only 8% of users click any result when an AI summary is shown, versus 15% without, and just 1% click a link inside the summary. If you walk into a QBR and lead with "AI sent us 340 sessions," someone will correctly observe that 340 sessions is nothing and cut the budget. What you should say is that 340 clicks, at roughly a 1% click-through from AI answers, implies your brand was read in AI answers a very large number of times, and that the sessions you did get convert far better than average. Measuring the click side properly in GA4 is worth doing. Leading with it is not.

The monthly slide

One slide. Four elements. In this order.

1. The chart. Share of Voice over time, you plus your top four competitors, last six months. Y-axis in percent, starting at zero. No dual axes, no stacked areas. If a competitor crossed you, the crossing point is the most important pixel on the slide and should be annotated.

2. The one number that changed. A single sentence at the top: "Share of Voice went from 15% to 19% this month, and we passed Cinder for the first time." Not five numbers. One, with its cause.

3. The gap table. Five rows, one per prompt archetype, three columns: your mention rate, best competitor's mention rate, delta. This is where the CMO's eyes will actually stop, because it's the only part of the slide that implies an action.

4. The ask. One decision. "We're at 8% on alternatives queries against Bolt's 61%. We need four comparison pages and three third-party placements. That's one writer for six weeks." AI visibility reporting that ends without an ask is a hobby.

What does not go on the slide: engine-by-engine breakdowns (appendix), the prompt list (appendix), screenshots of AI answers (one, maybe, if it's damaging enough to be a wake-up call), and any metric you can't explain in one sentence.

Cadence, and why weekly reporting to execs is a mistake

Measure weekly. Report monthly.

Weekly measurement is right for the team doing the work, because engine outputs drift and you want to catch real movement inside a sprint rather than a month later. But weekly exec reporting takes normal sampling variance and turns it into a fire drill. Your Share of Voice will bounce 2-3 points week to week for reasons that have nothing to do with anything you did. A CMO seeing that bounce every Monday will either learn to ignore the report or start asking you to explain noise. Both outcomes are bad.

Monthly is the right exec cadence. Quarterly is too slow, because AI engines change faster than that: Google AI Overviews expanded to ~48% of tracked queries, up 58% year over year, and ChatGPT referral traffic jumped 157.7% in a single week after one product change. A quarterly report would have missed both while they were still actionable.

Making the business case in one paragraph

If you need to justify the program itself, not just report on it, the case is short.

51% of B2B software buyers now start research in an AI chatbot, up from 29%. 69% chose a different vendor than the one they'd planned on, based on AI guidance, and 33% bought from a vendor they'd never heard of before the AI named it (G2, 1,076 decision-makers). The traffic that does arrive from AI converts at 4.4x the value of an organic search visitor. And the shortlist is being assembled in a place your existing analytics stack cannot see.

That's the argument. Not "AI is the future." Just: the shortlist is being written somewhere you're not measuring, and a third of buyers are choosing vendors they'd never heard of, on the strength of a paragraph you didn't write.

Building the report without a full-time analyst

The mechanical work, running 25 prompts across four engines multiple times a week in clean sessions, storing the responses, computing rates and Share of Voice, is exactly the kind of thing you should not be doing by hand, and exactly the kind of thing that quietly stops happening in month two when the person who owned the spreadsheet gets busy.

Spottlo produces the five KPIs above on a weekly cycle across ChatGPT, Perplexity, Gemini and Google AI Overviews, with all four engines on every plan. That last detail matters for this specific use case: most tools in the category gate engines behind higher tiers, and a per-engine KPI breakdown is impossible if you can only afford to see one engine.

What to do next

1. Delete the composite score from your current reporting, whatever tool produced it. If you can't reproduce the formula on a whiteboard, don't put it in front of a CMO.
2. Build the Share of Voice chart with your four real competitors on it, going back as far as your data allows. That's your slide.
3. Split mention rate by prompt archetype and find the row where you're furthest behind. That's your ask.
4. Add negative-mention rate to your tracking, and read the actual sentences. There is usually one that will change someone's mind about funding this.
5. Get a baseline scan and put a date on it. Every KPI on this list is a trendline, and a trendline needs a first point.