The diagnostic
Sample diagnostic report
One document in five sections, every figure carrying the cut it came from. The sample below is a real report on a real export: the pages read, the queries read, the search appearance breakdown, and a closing section on what the data does not single out.
What this sample is, and how it differs from yours
It is a real report on a real export, not a mock-up. The property is one we operate and is not named here, which is the rule every finding on this site follows. Its exports are not published, so you cannot check these inputs yourself; what you can check is that they agree with the findings already published from the same 28 days, which the build verifies on every deploy.
Two things will differ in yours. Your report carries your name and your site, where this one carries neither. And your queries file will be an interface export covering your whole property, capped at a thousand rows. The queries read here were pulled for one page cluster instead, which is why section 03 covers one family rather than all of them, and why the population it covers is stated rather than assumed.
The data read
- Window
- 28 days, 2026-08-13 to 2026-09-09, one pull
- Pages
- 440 exported rows, which become 382 canonical addresses carrying 2,079,623 impressions and 9,661 clicks
- Queries
- 3,328 rows, covering one page cluster rather than the whole property
- Search appearance
- 1 row
Addresses ending in a fragment are removed before anything is counted, 46 of them here carrying 29,647 impressions and no clicks, and www and trailing-slash variants of the same address are then merged. Both sides of every share below come from the rows that survive that, except where section 03 says otherwise.
01 Pages that answer in one line
Three groups of pages were separated by what their addresses identify them as stating: a branch code, an observance date, a calorie count. Each states one fixed value that a result can carry. That selection is made from addresses alone, not from click rates. The remaining 359 pages are the remainder and were not individually classified, so this is a comparison of three identified groups against everything else, not a review of all 382.
Calculators are in the remainder, which is a judgment rather than a measurement: a calculator produces a number from what the reader types, so the result page has no value to state. Someone could reasonably classify them the other way, and the figures below would change if they did.
| Group | Pages | Impressions | Clicks | Combined rate |
|---|---|---|---|---|
| State one value | 23 | 1,359,363 | 2,355 | 0.173% |
| Every other exported page | 359 | 720,260 | 7,306 | 1.0% |
| Bank branch codes | 16 | 803,667 | 1,774 | 0.221% |
| Festival and observance dates | 5 | 367,986 | 250 | 0.068% |
| Calorie lookups | 2 | 187,710 | 331 | 0.176% |
The three groups carry 65.4% of the impressions in the export. The remaining pages were clicked at 5.9 times their rate. That ratio is the reading; the cause is not in this export. Finding 01 groups the same export by subject rather than by what a page states and reports one page more: an explainer about what a bank code is, which belongs to the subject and states no code.
02 Seen often, clicked rarely
Pages ranked by impressions alone, clicks playing no part in the ranking, and each cut compared with every page below it on combined rate rather than on an average of page rates.
| Cut | Share of impressions | Their rate | The rest | Ratio |
|---|---|---|---|---|
| Top 5 | 64.9% | 0.180% | 0.989% | 5.5x |
| Top 10 | 80.7% | 0.189% | 1.6% | 8.6x |
| Top 20 | 88.2% | 0.284% | 1.8% | 6.4x |
Inside one position band
Ranking across the whole export mixes positions, so the same comparison was run again inside a single band. Of the 180 pages averaging position 5 to 10, the ten shown most were clicked at 0.189% and the other 170 at 1.8%.
Position is not held constant even there. The ten most-shown average 7.65 and the other 170 average 6.43, a gap of 1.22 positions running against the most-shown pages. So position cannot be dismissed as an explanation, and how much of the gap it accounts for is not established here. A gap between two group averages does not bound the effect of the positions underneath them.
03 Common and rare queries, within one family
The query rows returned for one page cluster: 3,328 queries carrying 636,317 impressions and 820 clicks.
How much that leaves out cannot be stated as one number, and the reason is worth the paragraph. The query rows were pulled by matching addresses on a text fragment of the path, which also matched 7 rows ending in a page fragment that the page count removes. So the pages behind the query rows carry either 810,421 impressions across 17 canonical addresses, or 833,763 across all 29 rows the filter matched, and the query file cannot show which of the two its own impressions belong to. The unreturned share is therefore 21.5% or 23.7% of impressions, not one figure. Clicks are 54.3% either way, because those fragment rows carry no clicks at all. Whatever is not returned, nothing here can say what it consists of, and no rate below is a rate for the cluster.
Within average position 8 to 10, so that position is roughly still, the 1,043 queries there were split into four groups of equal size by impressions.
| Group | Queries | Impressions | Clicks | Combined rate | Position |
|---|---|---|---|---|---|
| Most shown | 260 | 461,445 | 533 | 0.116% | 9.05 |
| Second | 260 | 10,752 | 16 | 0.149% | 8.94 |
| Third | 260 | 4,576 | 4 | 0.087% | 8.99 |
| Least shown | 263 | 1,207 | 1 | 0.083% | 8.81 |
Read these four rates as what they are: four observed ratios, not four estimates of an underlying one. The third and fourth groups rest on 4 clicks and 1, so one click either way would move them by a quarter and by double. The second rests on 16, which is also few. The ordering across the four is what the window produced, and nothing here establishes that the groups differ underneath, or that they do not.
04 Search appearance breakdown
Requesting the search appearance dimension over the same window returned one type, translated results: 184,133 impressions and 255 clicks, a rate of 0.138%, at average position 8.37.
No share of the property is stated for it, and that is deliberate. A share needs a denominator from the same population, and this is a property-level slice by appearance type with no address breakdown at all, so whether it includes the fragment rows the page count excludes is not knowable from the file. It must not be added to the page groups above either: it is a separate breakdown over the same impressions, so summing the two would count some of them twice.
05 What the exports do not single out
Said explicitly, because a report that lists only what looks actionable invites work the data does not support.
- 183 of the other 359 pages drew no clicks
- Most of them carry few impressions, where no click is an unremarkable outcome and one click would move the rate a long way. Whether any of them is worth attention is not something this export settles, and it is listed here so that it is not mistaken for a list of faults.
- The page rows add up to more than the property chart
- By 1.17 percent of impressions. That is how Search Console aggregates the two, not a defect and not a leak. It matters only because a share taken from one and divided by the other would mix two populations, so every share above uses the page rows on both sides.
- One country supplies 97.4% of impressions
- Across 196 country rows. That is context for reading everything above, not a finding: the rates here describe one market. It does not say the property should be in others, which this export cannot support either way.
- Nothing here is worth anything yet
- The exports carry impressions, clicks, click-through and position. They carry no revenue, no cost, no conversion and no session. A page with a low click rate may be worth more than one with a high rate, and nothing in these files can tell you which.
- Nothing here is a cause
- One window with no before and after shows how the groups performed, not what would change if anything moved. Every comparison above is between groups measured at the same time, and the one place position could be held roughly still is stated as roughly, not held.
How this report is checked
Not one figure above is written into this page. Every number is computed from the export by a script in the repository and read from the file it writes. Two of the three checks on that file run on every build: its own arithmetic, and its agreement with the findings already published from the same window. Where they overlap they agree, which is the canonical population, the top-ten cut, the within-band split, the appearance row and the query pull.
The third recomputes the file from the export and compares each source file with the checksum pinned beside it. That one needs the export, which sits outside this project, so it runs here before each release and on a deploy only where the export is reachable from it. The build says which of the three it ran, rather than reporting a check it skipped as one it made.
What that does not give you is independent verification. The export itself is not published, so the inputs are checked here rather than by you. About says the same thing about every figure on this site, and the method page sets out the rules each one follows. Applying the method to your own export tests whether the same pattern appears there; it does not verify these results.