In short: A report is a claim with its evidence attached. What separates it from an assertion is that somebody who was not there can follow any number back to what was asked, when, on which platform, and by whom.
Four things travel with every figure, and a figure missing one of them is a sentence rather than a finding.
Once it is sent it is a record of a date. Correct it by adding a correction, never by editing the page, because a report that changes quietly is worth less than one that was wrong in public.
A Claim And A Finding
"We improved our AI visibility by 30% this quarter" is a claim. Nothing in it can be checked, and a reader who wanted to check it would not know where to start.
The same result as a finding names the question set, the platform, the runs, the dates and the floor, and then says what moved. It is longer and less impressive and it is the only version that survives somebody asking how you know.
The test: could a colleague who has never seen your work reproduce the number from your report alone? Not get the same result, which would be surprising given how much these systems vary, but run the same measurement. If they would have to ask you a question first, the report is missing that answer.
Four Things Every Number Carries
The date it was measured. Not the date the report was written. These move underneath you, so a figure from three weeks before the deck was built is a figure about a different system.
The platform. Per platform, never blended. The overlap between what two assistants cite is small enough that an average across them describes no system that exists, which is the subject of The Role of Citations and Sources.
How many runs it came from. "Named in 3 of 5 runs" is a measurement. "We were named" is an anecdote, and the two look identical in a sentence.
Who ran it. Sessions carry history and accounts carry settings, so two people running the same set produce two different sheets and neither is wrong. Without this column a difference between quarters could be a difference between people.
The Shape Of The Page
Four blocks, and the reason for the order is that most readers stop after the first.
The decision you are asking for, in one sentence. The evidence, which is one comparison with its floor. What you shipped, with dates. Then the method and the limits, which the careful reader will reach and the rest will not need.
Putting the method first is the most common way to lose the room. It is the part that makes the finding defensible and it is not the part anyone is deciding on, which is the same ordering problem Goals That Survive a Meeting describes for targets.
Say The Limits Before Somebody Finds Them
Every measurement here has them, and naming them makes the covered part usable rather than suspect.
Ten questions is ten questions, not your market. One category does not generalise to another. Three runs is enough to beat a single reading and not enough to be precise. A quarter is one observation of something that moves on its own.
A reader who finds an unstated limit stops trusting the parts you did state. A reader handed the limits reads the rest as careful. It is the same mechanism that makes a page citable, and it works on people for the same reason.
The correction is the strongest part of a report, not the weakest. The study on this site published a control that was wrong by 14 points and said so. That correction is the most persuasive thing on the page, because it is evidence that a control was measured at all and that somebody checked it. A run of reports with no corrections in them is not a run of flawless quarters.
A Report Is A Record Of A Date
Once it has been sent, treat it as fixed.
If a number turns out to be wrong, issue a correction that says what it was, what it is, and what caused the error. Do not edit the original quietly. Somebody made a decision on the first version, and a document that silently becomes a different document makes every previous one unreliable too.
Keep them all, named by date, where the next person can find them. The value of a fourth quarterly report is almost entirely that the first three are still readable beside it.
Key Takeaways
- A claim cannot be checked; a finding names the questions, platform, runs, dates and floor, and is the only version that survives being questioned.
- Every number carries four things: when it was measured, which platform, how many runs, and who ran it.
- The page runs decision, evidence, work shipped, then method and limits, because most readers stop after the first block.
- State the limits yourself. A reader who finds an unstated one stops trusting the parts you did state.
- Never edit a sent report. Issue a correction beside it, and keep every version, because the comparison is the value.
Check yourself
Before you move on
Not scored, not recorded, and not part of the certificate. Both answers are settled by a sentence in this lesson, and the reasoning appears whichever option you pick.
- 01
Which four things have to travel with every number in a GEO report?
- 02
A number in a report you sent last month turns out to be wrong. What should you do?