Every figure in a defensible report carries four things: a metric with a definition, a named source, a date range including what it is being compared against, and any caveat that affects how it should be read. "Organic sessions, GA4, January to June, up 18 percent against the same period last year" can be checked by anyone. "Traffic up 40 percent" cannot, because traffic is not a metric and the comparison is missing.
The four things every number needs
| Element | Why | What it looks like missing |
|---|---|---|
| A defined metric | "Traffic" means four different things depending on who is asking | Traffic, visibility, engagement, reach |
| A named source | Two tools will give two answers for the same week, both correctly | A chart with no tool named anywhere on it |
| A date range | A percentage without a comparison period describes nothing | Up 40 percent, with no against-what |
| The caveat | Sampling, consent, definition changes, seasonality | A clean number hiding a messy one |
Metrics worth reporting
- Clicks and impressions from Search Console, for named query groups rather than site-wide.
- Organic sessions and conversions from analytics, with the conversion definition stated.
- Position on a named list of target terms, tracked consistently from the same locations.
- Indexation health, which matters and almost never appears in a report.
- Calls, form fills or bookings, which is what the business actually cares about.
- What shipped this period, because a month with no releases explains a month with no movement.
Metrics that fill pages and say nothing
- Site-wide average position, which moves whenever your query mix changes
- Total keywords ranking, which counts position ninety the same as position two
- Domain authority scores from third parties, which no search engine uses
- Bounce rate presented without context about what the page was for
- Impressions alone, presented as though they were visits
- Rankings gained, with no mention of whether anyone searches those terms
How to report a bad month
The bad months are the ones that determine whether anyone believes the good ones. A useful structure:
- State what happened, with the number, plainly and early.
- Say what you have ruled out and how. This is the part that demonstrates the work.
- Say what you believe caused it, and how confident you are.
- Say what you are doing about it, and when you will know whether it worked.
- Say what would change your mind about the diagnosis.
A report that only ever contains good news teaches the reader to discount all of it.
On things that cannot be measured well
Visibility in AI answers is the current example. The tooling is young, results vary between users and sessions, and nobody outside these companies can account for all of it.
The honest approach is to show what you sampled, how and when, and to label it as a sample. Presenting a sample as a measurement is the single fastest way to lose a client's trust, because eventually somebody checks.
Before sending a report, read each figure and ask whether you could answer "where did that come from" without opening anything. Anything that fails needs its source adding or needs removing.