We rank on page one and the click never comes

Position did not predict the click. Intent did.

The first thing to check when a page ranks and does not get clicked is the position, so that is where we started. We hold page one or two for 33 queries with at least 100 impressions between them, worth 12,661 impressions, and they earned not one click. At the click rate our own pages achieve at those positions, which is 0.57%, those impressions should have produced about 72 clicks.

The counter-example sits in the same export. A family of 52 queries that a buyer types when they already know what they want, ranked at a median position of 70.8, and converted at 26.4%. They produced 756 of the 918 clicks in the slice, which is 82.4% of every click we got, from pages nobody would call optimised.

So the variable that separates them is not the ranking, it is the shape of the question. A navigational query wants a page and clicks it. A query that asks for a comparison, a ranking or a recommendation wants an answer, and if the answer is already on the results page there is nothing left to click.

Our own research pages are the sharpest version of this. Across the export they hold 12,040 impressions and earned 7 clicks, a click-through rate of 0.06% against a site average of 0.87%. The pages are not the problem. The question they answer is one that people now ask an assistant, and the assistant is the thing taking the click.

Last verified 2026-10-22. This is one property over six months. It describes how this site behaved, and it is a starting point for your own numbers rather than a benchmark to hold others to.

What the export actually contains

The shape of the data decides what can be claimed from it, so it comes before the finding.

How the top thousand queries split

The slice, in two parts. The larger part is where the impressions are and the clicks are not.

906 of the 1,000 queries in the slice earned nothing, and they hold 79.7% of its impressions. This is the single most important shape in the export and it is not unusual.

The queries that earned impressions and no click

Every query in our export that ranked on page one or two, was shown at least 100 times, and earned no click. The wording is in the table because the wording is the evidence.

Some of these are not searches

Read the top row of that table again, slowly. It is not a question. It is a set of instructions.

The largest query in the set asks the search engine to compare HubSpot AEO with Scrunch, AthenaHQ, Profound and Otterly for brand visibility tracking across multiple AI engines, and then to provide a forced ranking from best to worst. It drew 1,344 impressions at position 3.03 and no click.

Four of the queries in the set are written in Spanish and one is written in German, and they all end the same way, with a demand for a forced ranking. Together the prompt-shaped queries hold 4,258 impressions, which is 33.6% of the set, between position 1.44 and 16.98.

This is what a query looks like when the searcher has copied a prompt into the search box. The person wanted a ranked opinion, and a ranked opinion is what an assistant gives them. Google served a page of blue links, the person took what they needed from the snippets, and the click never happened. Our page was in that list and it was not the thing they came for.

Three explanations, and the test that would separate them

The zero-click set has more than one possible cause, and the data can rule some of them out and not others.

The pages that do get clicked, and where they rank

The same export, the same window, the same site. These are the pages that earn clicks, and the position each one holds.

Each page, with what it earned

The six pages that hold the positions in the table above, one at a time. Two of them were written for a comparison query and two were written for a navigational one, and the difference in what they earned is the whole finding.

Click rate, against the position each page holds

The same six pages, from the table above, with the click-through rate each one earns. The bar for a page that ranks on page one and gets no click is drawn as an absence, because there is nothing to draw.

Click rate over the export window, from the page rows of the search console export of 22 October 2026. A bar is drawn for every page, and the pages with no click are marked as having none.

The AI-visibility cluster, on its own

The broader set contains a lot of noise, and one part of it is the part that matters. Here it is on its own.

The cluster is every query in the export containing aeo, ai visibility or ai search. That is 16 queries, holding 5,356 impressions, and they earned 0 clicks between them. Their positions run from 1.44 to 83.19, so most of them are on page one.

That is the category this site is built for. The comparison pages are written for exactly these questions, the questions are being asked, Google is showing our pages at the top of the results, and the click is not happening. There is no version of this where the ranking is the problem.

It is worth being clear that we cannot tell you what is happening on the results page, because the export does not say. We can tell you that the click did not happen, that the position was good, and that the queries are shaped like instructions to a machine. Everything past that is inference, and we have marked it as such.

What we changed, and how we will know

A measurement that changes nothing is a complaint with a chart on it.

Our research pages hold 12,040 impressions across 21 page rows in this export and earned 7 clicks, a click-through rate of 0.06% against a site average of 0.87%. The pages themselves are not the problem, and the answer is not to write them for a search engine instead of a reader.

So we split the work in two. Pages that answer a navigational question get the click, and we now treat a navigational query as a real target rather than a sidelined one. Pages that answer a comparison question get the citation, in an assistant and in a list, and we measure that in AI features rather than in clicks. The leading comparison page took 4,689 impressions in AI features alone, which is 71.1% of the AI-feature impressions in the export, and it recorded 496 page views in our own analytics over a longer window than Google reports.

The measurement plan is simple and we will publish it whether it works or not. We re-read the click-through rate of the comparison cluster at 30, 60 and 120 days. If it does not move, the honest reading is that comparison queries do not convert in Google, and that the citation is the return instead.

Four things the export says about the category

The findings, in the order they matter to a marketer looking at their own numbers.

What this looks like on a page you own

The abstract version of this study is easy to nod along to. Here is what it looks like when you look at your own numbers.

You pull your own query export. You sort it by impressions and you split it into two piles: the queries that earned a click and the queries that did not. On most sites the second pile is the big one, and it is holding most of the impressions, which is the first thing that feels wrong and is actually normal.

Then you pull the page export and do the same split. If the pages holding the impressions are the pages you wrote for comparison queries, and they earned nothing, you are in the situation this study describes. Ours held 12,040 impressions across 21 page rows for 7 clicks.

Then you do the bit that actually changes what you write. You take the zero-click page-one queries and you read them. Not the metrics, the wording. If they read like a question, you have a snippet problem and you can fix the snippet. If they read like an instruction to a machine, you have an intent problem and the fix is not a title tag.

Only after that is it worth asking whether the ranking is the problem. On our export it was not: 33 queries sat on page one or two and earned nothing, which is not a ranking problem at any position.

The five questions this study keeps getting asked

Four readers, and what each of them should take

The same measurement reads differently depending on what you are doing with it.

**You are writing for a query that names a tool, a format or a price.** You can win that query in Google. It has a destination, and the person wants to reach it.

**You are writing for a query that asks which of several things to choose.** You probably cannot win that query by being on page one. The person wants an opinion, and the opinion is the prize. Measure your return in citations and in the assistant's answer, not in the click, and stop treating the click-through rate as the score.

**You are auditing your own site.** The split between clicked and unclicked queries is the most useful cut in the export, and it takes five minutes.

**You are selling a tool that ranks in AI answers.** Your competitors' pages are sitting on page one earning nothing, and the space above the click is open.

What this does not tell you

Where this study sits in the rest of our work

Three studies that bear on the same question: what a page is actually for, and how you measure it.

How we checked

We analysed the Google Search Console export of 22 October 2026 and the GA4 page export for the same property. The daily series covers 183 days from 20 April to 19 October 2026 and gives the site's totals. The query export is the top 1,000 queries by impressions, which is a truncation the search console applies, and every share we publish is labelled as a share of that slice rather than of all traffic.

A query was counted in the zero-click set when it was shown at least 100 times, held a position between 1 and 20, and earned no click. The counterfactual is computed against our own rate for that position band, taken from the queries in the same band that did earn a click, so the comparison is between our pages and our pages rather than against a benchmark. The AI-visibility cluster is every query containing aeo, ai visibility or ai search, stated here so it can be re-run. The prompt-shaped subset is every query containing a forced-ranking phrase, and the four that qualify are shown in full on the page.

The page-level rows come from the page export, and the counter-example is chosen by clicks rather than by impressions, so that a page collecting large numbers of impressions and no clicks cannot stand in as evidence that a ranking converts. Every figure on this page is computed by one script from the exports, and the exports are hashed in the data file, so any number can be traced back to the row it came from.