In one sentence: a client report needs the method before the numbers, two rates rather than one score, openable evidence behind every figure, and an explicit record of what did not run — and the fourth is the one that makes the first three believable.
Lead with the method, in four sentences
The question set and its version. The engines and markets. The collection window. Whether brand-named questions were excluded and how competitors were detected. This costs a short paragraph and it pre-empts every question the client will otherwise ask three weeks later, usually in front of someone else.
A report that cannot state its method in four sentences is not measuring something stable enough to report.
Two rates, never one score
Mention rate and recommendation rate as separate columns. A brand mentioned in 40% of answers and recommended in 4% has a corroboration problem — work that happens off their domain. The reverse pattern is an attribution problem — work that happens on it. A single composite hides exactly the distinction that determines what you do next, and a client who later discovers the two numbers differ will reasonably ask what else was combined.
Citations: rank, not share
Report where the client's domain sits among all cited sources — "fifth of 109" — rather than a percentage. A share moves when the engine cites more or fewer sources overall, which has nothing to do with the work. Then show the source mix, because it is the most actionable thing in the report: it tells the client where the engine looks for their category, and therefore where the next quarter's effort goes.
Show the gaps
| Situation | Correct | What most reports do |
|---|---|---|
| Engine timed out | Blank with a note | 0%, which reads as a decline the client will ask about |
| No run that week | Gap in the line | Interpolate, which invents data |
| Question failed | Excluded, and said so | Dropped silently, raising the average |
Showing gaps feels like showing weakness. It does the opposite: a report with visible gaps is one whose other numbers can be trusted, and a client who sees you volunteer a gap stops auditing the rest.
What not to put in
- A rank number. Answers have no positions; a rank chart is order-of-appearance in one sample and it will move for no reason, which you will then have to explain.
- Month-on-month deltas on a single run. Two consecutive runs in the same direction, or it is not a finding.
- Any figure you cannot open. If clicking a number does not reach an answer, an engine and a date, leave the number out.
The cadence question
Report at the cadence you actually collect at, not monthly because that is the invoice cycle. If the quota supports one collection a month, a monthly report contains one data point against a metric that varies between runs — and saying that plainly is better than presenting a line drawn through twelve noisy points as a trend.