62,772 Impressions, Zero Clicks: What One Site's Own Search Data Says About Zero Click Searches

Everyone repeats that zero click searches are eating organic traffic. Almost nobody publishes the raw numbers from their own site, because the numbers are embarrassing.

Here are ours.

On 6 September 2026 we pulled every query secure-os.org ranked for, straight from the Bing Webmaster Tools API. The export covers 381 query rows, 68,596 impressions and 277 clicks, a sitewide click through rate of 0.40 percent.

130 of those rows carry 62,772 impressions and not a single click.

That is 91.5 percent of all the attention the site received, delivered and then not acted on.

The number that made us stop believing the obvious explanation

The easy story is position. We rank badly, therefore nobody clicks. We believed that for about ten minutes, until we split the export by average impression position.

Position clearly matters. But look at what else the table says: the two bottom rows hold 98.7 percent of the impressions and 25 percent of the clicks. The queries where we rank well barely have any volume, and the queries with volume are the ones nobody clicks.

If position were the whole answer, every query at position 8 would behave the same way. It does not.

The control group, which is the only reason this data is worth reading

A dataset that shows zero everywhere proves nothing, because the most likely explanation is a broken measurement. Ours does not show zero everywhere, and that is the point.

Here are real rows from the same export, same site, same week, same position band:

graphene os at position 8 converts. how to spot phishing emails at position 7, a better position, on seven times the impressions, converts nothing at all.

The clicks are being counted. The tracking works. Something is intercepting one kind of query and not the other.

The split that actually explains it

We tagged every row as a question or not a question. A question is a query starting with how, what, why, is, are, can, does, do, which, when, where or who, or containing a question mark. Nothing cleverer than that.

A query that is not a question earns clicks at 11.6 times the rate of one that is, on this site, in this export, at broadly comparable positions.

Fifty question rows carry 52,370 impressions between them and produced nothing.

That is the shape you would expect if the answer to the question is being rendered on the results page itself. We cannot see the results pages behind our own impressions, so we are not going to claim we watched an AI panel eat the click. We can only say that the pattern is the one a generative answer would produce, and that the two explanations we were able to test against this export both fail: position fails because better-placed questions convert worse than lower-placed entities, and broken tracking fails because the control group in the previous table records clicks normally. Seasonality, device mix and query intent drift are not tested here, and we are not claiming they are ruled out.

What we got wrong on the first pass, and why we are telling you

Our first look at this data used the top 20 rows the export tool printed by default. On those 20 rows, questions had zero clicks, and we very nearly published the sentence “questions produce no clicks at all.”

Pulling all 381 rows showed questions produce 61 clicks. Small, but not zero, and the difference matters: 0.116 percent is a bad channel, 0 percent would be a broken one, and those two conclusions lead to opposite decisions.

A truncated export is the most dangerous kind of wrong data, because it answers a slightly different question from the one you asked and it never errors out.

The rule this changed

We used to pick topics by asking whether a keyword had volume. how to spot phishing emails has enormous volume. It sent us 9,575 impressions in one week and zero people.

The question we ask now is different:

Not “does this query have volume”, but “can I realistically be in the top three, and will anyone still need to click when I get there?”

Both halves are required. Position 1 to 3 on this site converts at 29.4 percent. Position 8 on a question converts at essentially nothing, no matter how large the number next to it looks in a keyword tool.

Three practical consequences, in the order we applied them:

1. A question you cannot own outright is worth less than a niche entity you can. qubes os at 228 impressions returned more human beings than what is secure boot at 2,834.
2. Volume in a keyword tool is not demand for your page. It is demand for an answer, and the answer may never require your page.
3. Segment your own export before you trust anyone’s benchmark, including this one. Our numbers are one site, one search engine, one week. The method transfers. The percentages may not.

How to reproduce this on your own site

The export is free and takes about a minute:

1. Verify your site in Bing Webmaster Tools and get an API key.
2. Call GetQueryStats with the exact siteUrl string returned by GetUserSites. A URL you typed yourself returns an empty list with no error, which is its own trap.
3. Do not truncate the result. Sort by impressions descending, then compute click through rate per row.
4. Tag each row question or not question, and group by average impression position.
5. Look for your control group. If nothing in your export earns clicks, stop and fix your measurement before you draw any conclusion at all.

Bing returns one row per query per reporting period, which is why the same string can appear several times in the table above. We left the rows as the API returned them rather than merging them, so that anyone repeating this gets the same shape we did. Across the 381 rows there are 312 distinct query strings.

Anyone is welcome to reuse these figures. They are our own site’s data, measured on 6 September 2026, and we would rather they were quoted with the control group attached than repeated as another round number about the death of organic search.

The tables on this page are released under Creative Commons Attribution 4.0. Attribution is the only condition.

Cite this data:

Secure-OS, 62,772 impressions and the zero click rate of one real site, Bing Webmaster Tools export, 6 September 2026. https://secure-os.org/articles/zero-click-searches-data