Reporting AI Visibility to a CMO Who Didn't Ask For It

Most AEO programmes do not fail on execution. They fail at the point where someone has to explain, to a person managing a budget across six channels, why a new metric deserves attention and what it means when it moves.

That conversation has a predictable shape and a small number of ways to lose it. It is worth preparing for properly, because the reporting problem is upstream of everything else: a programme nobody understands does not get a second quarter.

Start before you need to

The single highest-leverage decision is timing, and it is usually made badly.

Introducing AI visibility reporting while organic traffic is stable is a strategic addition. Introducing it the quarter traffic falls is an excuse — and it will be received as one regardless of being correct. The content is identical; the reception is not.

If you are reading this while your numbers are fine, that is the moment. If you are reading it during a decline, say so explicitly and lead with the decline rather than the new metric, because attempting to reframe a bad quarter with an unfamiliar chart is how a metric gets permanently associated with excuse-making.

The one slide

Executive attention for a new metric is roughly one slide and ninety seconds. Four things earn their place.

A competitive gap, stated as a share. “We appear in 40% of the answers that matter to our buyers; our closest competitor appears in 70%.” This is the most useful sentence available because it contains a target, a trend and an implied action, and because it is relative — it survives the question “is 40% good?” which no absolute figure does. Share of voice explained is the formalisation.

Direction over time, not a delta. Five or six points on a line. A single arrow between two scans is sampling noise, and presenting it as movement is the fastest way to be caught out by someone who checks.

One representative answer, quoted verbatim. The actual text an engine returned when asked a question a real buyer asks. Nothing else in the deck does what this does — a paragraph naming two competitors and not you makes the abstraction concrete in a way no chart manages. This is the slide element people remember and repeat.

What you are doing about it, in one line.

That is the whole slide. Everything else is backup.

What to leave off

Anything implying revenue attribution. You cannot connect an AI mention to a purchase — the chain routinely runs through a direct visit weeks later with no identifier surviving. A modelled revenue figure will be challenged, you will not be able to defend the assumptions, and the challenge will discredit the parts of your reporting that were sound. What AEO cannot do is the full list of things not to claim.

Small movements. A mention rate over a hundred observations carries roughly a ten-point margin. Reporting a five-point change as progress is reporting a random draw, and it sets an expectation you will have to meet next month with another random draw. How many runs before you trust an AI visibility number has the arithmetic — worth having in backup, because the first sharp CMO will ask.

Per-engine detail. Interesting to you, noise to them. It goes in the appendix and comes out only if asked.

Prompt volume. It does not exist. If a vendor dashboard displays one, do not put it in front of an executive, because it cannot survive “where does that number come from?”

The four questions, and honest answers

“Is this actually driving revenue?” No, and nobody can show you that it is — the attribution does not exist for anyone in this category. What we can show is that our buyers are asking these questions, that our competitors are named more often than we are, and that the gap moves when we work on it. That is the same evidentiary basis brand advertising has always run on.

Answering this one honestly is the highest-stakes moment in the whole conversation. A confident fabricated number wins the meeting and loses the programme in Q3.

“How big is this really?” Growing and not yet primary for most categories. The honest framing is that it is a leading indicator: the questions being asked are the same ones that used to enter through search, and the share of them resolved without a click is rising. Pew Research’s 2026 survey on Americans and AI is a reasonable external reference point for adoption, and it is not a substitute for measuring your own category.

“Why is traffic down if this is working?” Because they are different things, and this is worth separating carefully. Some traffic decline is answers satisfying informational queries that were never going to convert — a loss of low-intent visits. Some is competitive or seasonal and has nothing to do with AI. Report the composition: if sessions fell but conversion rate rose and branded search grew, the visits you lost were the cheap ones. AI search and the end of the session metric develops the argument.

“Can’t we just pay to be included?” No. There is no ad placement inside the recommendation itself, and anyone offering one is misrepresenting an ad product.

Set the cadence expectation early

The mechanism does not produce monthly-legible movement. Grounded answers respond to content in days to weeks; third-party consensus shifts over months; training-derived knowledge changes across model generations.

So agree the reporting rhythm at the start: activity monthly, outcome quarterly. A monthly outcome report will show noise, the noise will occasionally look like a decline, and you will spend the meeting explaining variance instead of the work. How long does AEO take is a useful thing to circulate before anyone forms an expectation you cannot meet.

One more thing to flag pre-emptively: a model update can move the whole category overnight with no action by anyone. Saying that in month one costs nothing. Saying it in month six, the day it happens, sounds exactly like an excuse.

The counter-argument

The reasonable objection: this is a lot of hedging, and an executive who hears “we can’t attribute revenue, movements under ten points are noise, and results take a quarter” will conclude the whole thing is unmeasurable and defund it.

That risk is real and the mitigation is not to soften the caveats — it is to lead with the strongest true claim rather than with the limitations. The strongest true claim is the competitive gap, and it is genuinely strong: it is directly observed, it is specific to your buyers, and it moves in response to work. Open with that, and let the caveats arrive as answers to questions rather than as a preamble.

The failure mode this article is guarding against is the opposite one, and it is more common: a programme that overclaims early, gets a good first quarter of attention, and is quietly killed when someone senior asks for the revenue line and finds nothing behind it.

Where to start

1. Build the one slide before you build anything else. If it does not hold together, more data will not fix it.
2. Pull one verbatim answer that names competitors and not you. It will do more work than the chart.
3. Agree cadence explicitly: activity monthly, outcome quarterly.
4. Write down what you will not claim, and say it out loud in the first meeting.

AEO for marketing managers covers the reporting line, and AI visibility reporting template has the structure if you want a starting document.