What a tracking report should say, and refuse to say.
Most reports are a wall of numbers that answer a question nobody asked. Four sections cover what a business actually needs, and one rule keeps the whole thing honest.
What should a monthly AI visibility report contain?
The test is simple and unforgiving: could the person receiving it decide something? A report that produces no decision has cost the reader time and told them nothing they can act on, however many charts it contains.
Four sections
- What changed. Only movements outside normal variation, each with its source and its date. Everything that moved within its usual range is deliberately absent, and the report should say that it was checked and was stable, because restraint that is invisible reads as work not done.
- What it means. The interpretation, in a sentence, including the cases where the honest interpretation is that a change is real and does not matter yet.
- What is being done, and by whom. Every finding either has an owner and a date or is explicitly parked. A finding with no owner is a rumor, and a list of findings with no owners is how a report becomes something nobody opens.
- What happened to last month's actions. The section that makes the previous three worth reading. Work completed, and whether the signal it targeted actually moved, including when it did not.
What it must refuse
- Precision it did not measure. No confidence figure attached to a sample of one, no percentage derived from a number that was itself an estimate, no trend line through two points.
- A blank where a check failed. If a source could not be reached, the report says so. Unmeasured is not zero, and a missing reading rendered as a decline sends somebody to fix a problem that does not exist.
- A metric that only ever goes up. If no plausible month produces a worse number, the number is decoration.
- Anything the business cannot act on. An interesting fact about the category with no available action belongs in a conversation, not in an accountability document.
The uncomfortable version is the useful one
A report where the honest answer is "the foundation held, two things need attention, and the change we made last month did not move what we expected" is worth more than one showing eight green metrics. It is also the harder one to send, which is exactly why it is worth checking whether your reporting is capable of producing it.
If it is not, the reporting is not measuring. It is reassuring, and it will keep reassuring right up until something matters.
Who the report is for
An owner and an operator need different documents from the same measurements. The owner needs to know whether the position is improving, what it is costing, and what needs a decision. The operator needs the specific queue: which page, which profile, which line, in what order.
Collapsing them produces the familiar artifact that satisfies neither, where an owner scrolls past technical detail to find nothing decidable and an operator gets a summary with no instructions. It is better to write the short accountability document and keep the working queue where the work happens, each linking to the other.
This is one question beneath AI visibility tracking, the ongoing work Digilu does through The Observatory. The free point-in-time baseline is AIOInsights. Digilu cannot make a private AI model recommend a business. It can make the public evidence clearer, stronger and easier to verify, then track whether visibility improves.