Home › Tools › The Reporting Integrity Check
Free check · 7 items · 4 minutesCan your monthly report tell a surface change from a content failure?
Seven items, scored on the noes. If the report merges behaviours that move independently, every explanation built on it is a guess — including the ones that get pages rewritten.
How do you check an SEO report?
Check an SEO report by testing whether it separates behaviours that move independently. Confirm that impressions and clicks are separate lines, server logs are available, AI traffic is split into three buckets, every number carries a confidence tier, AI visibility is reported per engine, citations sit beside clicks, and CTR is decomposed before anyone explains it.
- Impressions and clicks as two lines. Each with its own owner and its own explanation, because they now move independently.
- Server-log access covering at least 30 days. Client-side analytics record no AI agent fetches; those requests exist in logs or nowhere.
- Three AI buckets, not one. AI agent fetches, AI search referrals and agent workflow hops reported separately, never as one “AI traffic” figure.
- A confidence tier on every number. Observed, inferred or panel-derived, so a reader knows what was measured and what was estimated.
- AI visibility per engine. One line per engine, with any excluded series named and the exclusion justified; a blended average describes an engine nobody uses.
- A citation line next to the click line. Without it, a page that is quoted without being visited is invisible.
- CTR decomposed before it is explained. Check whether impressions moved before clicks did; CTR is a ratio, and both halves have to be read first.
Frequently asked questions
How do you read an SEO report?
Read an SEO report from the denominators up. Check impressions before clicks, because CTR is a ratio and either half can move it. Then ask what each AI number counts, on which engine, and whether it is observed, inferred or panel-derived. A number with no tier and no engine is a claim, not a measurement. Read the full method.
Our own numbers, measured on our own estate
We scored Merged the first time we ran this on ourselves — items 1, 3 and 4 all came back no. That is the honest baseline.
| Measured | What it is |
|---|---|
| at least 543 queries / 7,038 impressions | Page-one queries that earned zero clicks in 28 daysEight queries earned any click, and all eight were brand or navigational. "At least" is deliberate: the export caps at 500 rows per slice, so this is a floor, not a census. · Indexable, 28-day Search Console window to 6 September 2026 |
| 55.7% of URLs, 6.4% of clicks | Share of our estate in substitutable content typesClick share alone never reveals what is being quoted without being visited. · Indexable, September 2026, n = 230 pages |
| 68.01% | US searches that ended without a clickJanuary to April 2026. Third-party market context, not our estate. · SparkToro, 2026 |
| zero | AI agent fetches recorded by client-side analyticsAgent requests do not execute your JavaScript. They exist in server logs or they exist nowhere. · Osmani, 2026 |
One domain, one window. Every figure here carries its own confidence tier: the zero-click counts are observed and the query-attribution reading is inferred. A case study, not a law.
A report that cannot distinguish two behaviours will get a working page rewritten
The reporting split is one afternoon of Search Console work and one conversation about who owns which line. The citation line needs an instrument — that is the part an audit builds for you.