An AI visibility report earns its place when it tells the client which buyer questions they were named on, which they were missing from, and which third-party pages the engines cited instead. The single headline score is the least useful number in it. Across eight weekly readings of our own brand, the score moved between 5.7 and 8.5 with nothing shipped to move it.
Disclosure: GetIntel is an AI visibility tool that sells client reporting to agencies. Every first-party figure below is our own measured data, with the window it covers stated.
What can you actually measure in an AI visibility report?
Four things are measurable and worth reporting: how often the engines name the client in their category, which specific buyer questions produce a mention, which domains get cited instead, and how those change over time. GetIntel measures this by running a fixed set of buyer questions across ChatGPT, Perplexity, Gemini and Google AI Overviews on a daily schedule, then scoring on a rolling seven-day window rather than a single day's answers.
Two things are not measurable, and promising them to a client is where agency reporting goes wrong. Nobody can attribute a specific sale to a specific AI answer, because there is no referrer on a conversation the buyer had before they ever reached the site. And nobody can tell a client why an engine changed its mind, because the engines do not publish their retrieval logic.
Why should you not report a single score to a client?
A single AI visibility score is unstable enough that reporting it alone will mislead a client in both directions. Here is our own brand's score, measured across 60 tracked buyer questions on four engines, read on 20 September 2026:
| Week ending | Score |
|---|---|
| 3 August 2026 | 7.3 |
| 10 August 2026 | 6.0 |
| 17 August 2026 | 8.5 |
| 24 August 2026 | 5.7 |
| 31 August 2026 | 7.3 |
| 7 September 2026 | 6.5 |
| 14 September 2026 | 6.7 |
| 20 September 2026 | 6.8 |
The mean of those eight readings is 6.85 and the median is 6.75. The high is 49 percent above the low. The single week from 17 to 24 August fell 2.8 points, a 33 percent drop, and we published nothing in that week aimed at moving it.
An agency reporting on a mid-August cycle would have handed the client an 8.5 and taken credit. The same agency reporting a week later would have had to explain a third of it disappearing. Neither conversation would have been about anything real. The wider problem with treating one number as the verdict is covered in why AI visibility scores can be misleading, and the underlying instability is measured in how many reruns it takes before an AI visibility number settles.
The four blocks of a monthly client report
A monthly client report should lead with prompt-level detail and put the score last. Four blocks cover the job:
| Block | What it answers | Why it survives scrutiny |
|---|---|---|
| Questions named on | Which buyer questions produced a mention | The client can verify it by asking the engine |
| Open questions | Which tracked questions returned no mention | Names the next month's work |
| Cited sources | Which domains the engines quoted instead | Points at where placement is needed |
| Movement | What changed, and whether it exceeds normal noise | Stops a noisy week reading as a result |
Our own current reading covers 60 tracked questions across four engines, drawn from 1,454 answers in the seven-day window ending 20 September 2026, and puts our share of voice at 2.2 percent of the citations in our category. The share of voice figure is the one worth putting in front of a client alongside the prompt list, because it is a ratio against named competitors rather than an absolute. Measuring share of voice sets out how that ratio is built and what denominator it uses.
Can an agency put its own branding on the report?
Branded reporting exists on both of GetIntel's price lists, and which one an agency uses depends on how it bills.
| Plan | Price | Covers | Branded and scheduled reports |
|---|---|---|---|
| Pro, self-serve | $79 / month | one brand | included |
| Agency | $59 / client / month | 10 to 24 client brands | white-label and scheduled client reports |
| Partner | custom | 25+ client brands | white-label partnership terms |
The rate card also has an entry-level Studio tier at $79 a client brand for three to nine clients, the starting point for agencies with at least three clients.
Agencies comparing branded reporting across vendors will find the feature described very differently from one product to the next, which is why the white-label AI visibility tools comparison states what each one actually rebrands rather than whether the phrase appears on the pricing page.
How do you run reporting across a dozen clients?
Running AI visibility reporting across a dozen clients is a scheduling problem, not a per-client analysis problem. Three mechanics do the work. Scheduled reports remove the monthly assembly step. A Looker Studio connector pulls the whole roster through one key rather than one export per client. Per-client billing means the roster grows without a plan migration, because the rate card scales by client count rather than by seat.
What does not scale is the interpretation. A report that goes out unread is worse than no report, because the first time a client asks why their score fell the agency has to admit nobody looked. Budget the reading time, not just the generation.
How do you prove the retainer is working?
Prove an AI visibility retainer with the count of buyer questions where the client is named, not with the score. A question that flipped from absent to named is a concrete event: the client can open ChatGPT, ask it, and see their own name. A score moving from 6.5 to 6.8 is not something a client can verify, and the same week-to-week swing that took our own reading from 8.5 on 17 August to 5.7 on 24 August makes a small score move an unreliable signal in its own right.
GetIntel reports which of the tracked questions do not yet name the brand for exactly this reason. That count falling is the clearest evidence the work is landing, and it survives the week-to-week noise, because a single question does not flip back and forth the way a composite average does.
Pricing the reporting inside the retainer
Reporting should not be a separate line item on the retainer, because the client is buying the outcome rather than the PDF. What matters is that the reporting cost sits inside the margin rather than eating it. Two independent agency-pricing sources put a real AI visibility retainer for a small to mid-market client at $1,500 to $8,000 a month. Against that range, a per-brand software cost of $59 to $79 works out between roughly one and five percent of what the client is billed, which leaves the retainer's economics intact whether the agency reports monthly or weekly. The real margin math on an AI visibility retainer works that gap through with independently sourced retainer figures rather than vendor claims.
The shortest version
Report what the client can check. Buyer questions named and missing, the domains cited instead, and the direction of travel with an honest note on noise. Put the score at the bottom, state the window it covers, and never let a single week of it become the story. Our own number moved 49 percent across eight weeks without a single page being published to move it, and any agency reporting monthly will see the same thing on their clients' numbers.