18,232 Impressions, Zero Clicks: What Our Search Data Actually Shows About Zero-Click Searches

We run a network of sixteen sites and read their search data every day. On 5 September 2026, on one of them, two queries produced 18,232 Bing impressions and not a single click between them. They sat at average positions 7 and 8, at the bottom of the first page.

That figure is easy to misread. The usual reaction is to conclude "our titles are bad" and rewrite them. We did the opposite: we first checked whether this zero was real.

The raw numbers

Three rows from the same site, over the same period:

The third row is the most important of the three, and it is the one nobody looks at.

The control that makes the zero credible

A broken counter returns zero. A working counter also returns zero, when there is nothing to count. The two look exactly alike, which is why a zero on its own never proves anything.

Query C is our control. One impression, one click, recorded. The click counter on this site therefore works, and it proved it at the very moment the two other queries stayed at zero. Without this row, we would have spent hours looking for a measurement fault that did not exist.

The method carries over to any measurement: before concluding from an absence, find in the same dataset an occurrence that was recorded.

What zero-click searches really mean here

Position 7 or 8 used to mean "few clicks". Our figures describe something else: not a low rate, but a threshold. At position 1, one impression out of one produced a click. At positions 7 and 8, eighteen thousand two hundred and thirty-two impressions produced none.

This is not a difference of degree. The top of the page is taken by generated answers, rich snippets and blocks that answer the question without a link being needed. The link ranked seventh is no less attractive than before; it is simply out of view for someone who already has their answer.

The keyword rule we drew from it

We changed how we pick a topic. The criterion is no longer "does this keyword have volume?" but "can I rank in the top three?".

The consequence is counter-intuitive and can be put in numbers: two hundred impressions you can reach at position 1 are worth more than twenty thousand impressions low on page one. The latter bring in zero, and zero times the volume is still zero.

In practice, this made us drop short, generic queries in favour of long, precise ones: exact error messages, questions that carry a specific constraint. Competition there is low and the intent is unambiguous.

Three caveats, because a figure without its limits is a slogan

The takeaway

A zero is checked before it is interpreted, and it is checked with a control taken from the same dataset. Once established, it is worth more than an average: it marks a boundary, and a boundary is worked around rather than optimised.

If you track a site whose impressions rise while clicks do not follow, the first question is not the title. It is the real position on the queries that carry the volume. The same logic shapes how we write a website requirements document: decide the expected action first.

Reusing these figures

These measurements are free to reuse, including commercially, provided you cite the source and keep the reading date: a search figure is a snapshot in a sliding window, and it does not mean the same thing three months later.

If you replay the control method on your own data and get something different, that interests us more than a confirmation: a reading that does not reproduce on another account says something ours cannot.