Most enterprise AI-visibility programs are not cancelled because they failed. They are cancelled because nobody could say clearly what they achieved.

The pattern is consistent enough to plan around: good work happens, a number goes up, and at the second budget review someone asks what that number means for pipeline. There is no answer, because the number was never designed to have one. The program ends, and it ends looking like the topic did not matter — which is the expensive outcome, because the next attempt starts from a worse position internally.

So this is the reporting-first version. What to measure, who owns it, and what to put in the first ninety days so the review goes well.

Report three numbers, not one

Almost every tool in this space gives you one visibility percentage. Do not report it. It bundles three things that move at different speeds and belong to different people, and bundling them is what makes the review unanswerable.

Number 1 — Presence rate

What it is: of your category questions, the share where your brand is named at all.

Who owns it: communications and industry relations. It is driven by how widely third parties describe you, not by your website.

How fast it moves: quarters.

The catch: measure it on category questions, never on your own brand name. "What does Acme do" always names Acme and tells you nothing. "Which vendors handle X for Y" is the buyer's question and the only one worth counting.

Number 2 — Description accuracy

What it is: when you are named, does the description match what you actually sell, to whom, today?

Who owns it: product marketing, with web. This is your own material talking.

How fast it moves: weeks. This is your quick win, and it is where most enterprises have the biggest gap without knowing it.

What you will find: the description is usually of the company you were five years ago, because five-year-old material is plentiful and current material is thin. Divested products still attributed to you. The wrong customer segment, borrowed from whichever case study got the most coverage.

Number 3 — Shortlist inclusion

What it is: on recommendation-shaped questions — "which vendors should I consider" — are you on the list?

Who owns it: product marketing and sales together, because it depends on published comparative evidence.

How fast it moves: slowest of the three.

Why it is the one that matters: this is the number that maps to pipeline. It is also the one least correlated with presence — you can be named everywhere and recommended nowhere. Mechanism in how AI engines decide what to recommend.

The reporting mistake that ends programs

Averaging the three, or reporting only presence.

Presence is the easiest to move and the first to rise. Report a composite and it goes up early, leadership is pleased, and expectations get set against a number that has already saturated. Two quarters later presence is flat because it has nowhere left to go, inclusion never moved because nobody was working on it, and the program looks stalled.

Three separate numbers, each with an owner, each with its own expected speed. When presence saturates, that reads as a stage completed rather than as momentum lost.

What to put in the first quarter

Weeks 1–2: Baseline

Budget about two days. Do not buy a tool for this yet; the coding judgement is the valuable part and you want to understand it before you outsource it.

Weeks 3–6: Fix description accuracy

This is where the fast return is. Take your ten highest-authority pages and convert positioning language into structured facts — replace "industry-leading solutions" and "two decades of commitment to quality" with what you sell, to whom, at what scale, with what evidence. Step-by-step version in the eight-step checklist.

Get legal involved in week 3, not week 6. Specific claims need review, and discovering that in week 6 costs you the quarter.

Weeks 7–10: Publish comparative evidence

For shortlist inclusion, systems need a reason to place you. Publish the conditions under which you are the right fit — segment, scale, use case, integration, jurisdiction — in plain prose on a page a crawler reads.

Include what you are not for. Counterintuitive and consistent: a stated boundary makes you easier to recommend confidently, because it gives the system a condition to match against.

Weeks 11–12: Re-measure and report

Same 30 questions, same platforms, same coding. Report the three numbers separately with the owner named against each, plus three verbatim before-and-after quotes from the transcripts.

The quotes do more work than the numbers. A slide showing what an AI system said about your company in January next to what it says in April is the most persuasive artifact this work produces.

Who needs to be in the room

The reason enterprise programs stall is rarely disagreement. It is that the work sits across four functions and belongs to none of them:

Name one owner and one executive sponsor in the same meeting where the program is approved. Sequencing detail in the 90-day rollout plan.

Common questions

Should we buy a visibility tracking tool?

Not in the first quarter. Run it manually once so you understand what the coding decisions actually are — most tools measure presence well and description accuracy not at all, and you cannot evaluate that until you have done it by hand.

How many questions is enough?

Thirty is a working minimum for an enterprise. Below twenty, normal week-to-week variation swamps any real change and you will spend the review explaining noise.

How much variance is normal?

Enough that a single run is not a reading. Same question, same platform, different day gives different answers. This is why the question set has to be fixed and the cadence regular — see our 60-day tracking data for what the fluctuation actually looks like over time.

We are a large company with a long history. Does that help?

It helps presence and hurts accuracy. Lots of material means you are easy to find and easy to describe out of date. Large enterprises usually have their biggest gap in number two, and it is usually the cheapest to close.

What if leadership only wants one number?

Give them shortlist inclusion and show the other two as the drivers behind it. It is the one closest to revenue, so if you are being held to a single figure it should be that one — not the flattering one.

Where does this framework come from?

The three-way split is the measurement side of the staged model in GEO: The New Enterprise Gateway in the AI Era — discovered, understood, shortlisted. The book's contribution is the diagnosis; this is what it implies for reporting. Its CEO self-audit checklist is also a faster first read than a full measurement cycle. Notes in our summary, wider shelf in the reading list.

Book details

Further reading

See also what to check when your brand stops appearing, which chapter to give each executive, and the ChatGPT visibility guide.