Seventy seven percent of our impressions could never have become a visit

Impressions are the metric everyone reports and almost nobody audits. They get treated as latent demand: people who saw you and could have clicked, waiting to be converted by a better title or a higher position. On the property below, three quarters of that pool was not latent demand. It was demand already satisfied on the results page, before a click was possible.

The size of it

Over twenty eight days the property returned 2,079,623 impressions and 9,661 clicks across 382 pages, a rate of 0.465 percent. Four clusters account for 76.7 percent of every impression the site earned, all of them on page one, and they convert at 0.177 percent.

One pull, one window, page level throughout, so the clusters are disjoint and the shares add up. URLs are canonicalised first: fragment rows removed, then www and trailing-slash variants merged. The denominator is the sum of these page rows, not the property day chart, which totals 1.2 percent less and would put a numerator and denominator from two different aggregations in the same fraction. The search-appearance dimension is deliberately excluded: it describes the same impressions from another angle.

The ratio that matters is not 77 percent

It is 0.177 against 1.413. The part of the site that can convert does so eight times better than the part that cannot, on under a quarter of the impressions. Every site-wide average this property has reported has been a blend of those two, weighted three to one toward the half that was never going to produce a visit.

What makes an impression impossible

The test is the shape of the answer, not the quality of the page. If the query resolves to a single value, the engine can render the whole answer in the result and the searcher is finished. A branch code is one string. A festival date is one date. A calorie count and a fuel price are each one number.

Notice what this is not. It is not thin content, a slow page, or a weak title. The pages inside that 76.7 percent were built to the same standard as the ones outside it, by the same people, on the same template. The difference is entirely in what the searcher wanted.

Where the number is sensitive, stated plainly

The headline moves with the classification, so here is the rule and the arithmetic rather than the number alone. On bank codes and calorie lookups only, the two clusters this finding originally covered, the share is 48.0 percent. Adding festival dates and fuel prices takes it to 76.7 percent.

Both additions resolve to a single value and both convert below 0.2 percent, which is why we include them. A reader who disagrees can recompute from the table above, which is the point of publishing it.

Two earlier versions of this number were wrong

The first said 43 percent, computed against a denominator from a different pull. The second said 79 percent, computed against the property day-chart total while the numerator came from the page export. Page rows sum to 1.2 percent more than the day chart, so that fraction mixed two aggregations.

Neither error was large enough to change the argument, which is exactly why they survived. The rule that catches them is duller than the finding: a numerator and a denominator have to come from the same count of the same thing.

Why the aggregate sets the wrong target

Anyone briefed to lift 0.465 percent goes after the biggest clusters first, because that is where the impressions are. On this site the biggest clusters are precisely the ones with no click to win. The brief sends the work to the one place it cannot pay.

How to check your own

Export Search Console by page for a full month. Strip fragment rows, merge www and trailing-slash variants, and only then count. For each page with meaningful impressions, ask one question: does the query it ranks for resolve to a single value? Sum impressions where it does and divide by the sum of the same rows, never by a total from a different report.