We Blamed AI Overviews. It Was Our Titles.
Nine hundred impressions at positions three to ten, and zero clicks. We assumed zero-click search. The Search Console data said something duller.
For sixteen months this site accumulated impressions and almost no clicks. In the most recent twenty-eight-day window, the blog earned 989 impressions and zero clicks.
The interesting part is not the number. It is that several pages sit in the top ten for the queries producing those impressions, and still get nothing. One page ranks third and has never been clicked.
We assumed zero-click search. That assumption was mostly wrong, and the real explanation was duller and entirely our own fault.
What the data showed
Pulling non-brand queries where the site ranks in the top twelve:
| Query | Position | Impressions | Clicks |
|---|---|---|---|
| best automation tools for fortune 500 companies' mission-critical processes | 6 | 169 | 0 |
| best automation tools for fortune 500 companies | 7 | 151 | 0 |
| what tasks are the biggest time sinks that ai can reduce | 8 | 108 | 0 |
| what are the tools helpful for fortune 500 technology needs | 3 | 76 | 0 |
| mitigating ai risks strategies for cisos 2025 | 4 | 56 | 0 |
Roughly nine hundred impressions across five pages, at positions that should convert, producing nothing at all.
The hypothesis that felt obvious
These are long, conversational, question-shaped queries. That is exactly the profile that triggers a generated answer at the top of the results, and a generated answer that satisfies the searcher removes the click.
The story was coherent, it matched the industry conversation, and it had the additional appeal of being nobody's fault. We were ready to accept it.
What we actually found
Before accepting it we looked at what the pages display in the results, which we should have done first.
The page ranking fifth to seventh for "best automation tools for fortune 500 companies" was showing this title:
7 AI Productivity Tools Fortune 500 Teams Use by 8 AM
Someone searching for enterprise automation tooling sees a listicle about reclaiming your morning. The page is genuinely relevant, its body is a structured breakdown of seven categories of automation tooling, which is why it ranks at all. But nobody clicks, because the title says it is about something else.
Every one of the five pages had the same defect. They rank on body content and lose the click on framing.
Then we checked the meta descriptions, and found a second problem sitting underneath the first. 61 of 70 legacy posts had a description that stopped mid-sentence:
By leveraging cutting-edge artificial intelligence, these leading enterprises are not just working harder, they're working smarter. They
A generator had truncated the opening paragraph at a character limit and written the fragment into the field. So the searcher saw a mismatched title above a sentence that breaks off unresolved. Both halves of the result were working against the click.
Why we did not catch it
The audit tooling had been reporting these as fine, and it was right to. The descriptions were the correct length, unique, and present. Nothing checks whether a description ends mid-clause, and nothing compares a page's title against the queries it actually ranks for, because that comparison requires joining Search Console data to the content, which no standard audit does.
The failure was not that a check failed. It was that the check did not exist, which is the harder kind to find — the same shape of problem as an agent pipeline where every component reports success.
What we changed
Two things, both cheap.
Titles on the five affected pages were rewritten to match the intent of the queries they already rank for, keeping the terms the data shows are working. The body content was not touched, because the body content was never the problem.
The 61 truncated descriptions were removed entirely, falling back to hand-written descriptions that are complete sentences within the length Google displays. Deleting a broken field beat writing sixty-one new ones.
Neither change touches what the page says. Both change only what the result looks like.
What we still do not know
We cannot separate the two effects, and we are not going to pretend otherwise. Search Console does not report whether an impression appeared alongside an AI Overview, so absorption and intent mismatch produce identical data. It is entirely possible that both are happening and we have fixed the half we could see.
We also do not know how much of the ranking rests on the current titles. Titles are a ranking input as well as a click input, and a rewrite can cost position. We kept the working terms to limit that risk, and we will find out.
The honest summary is that we found a sufficient explanation, not a proven one. What made it worth acting on is that it was cheaper to test than the alternative, and the alternative was unfalsifiable with the data available.
The transferable part
Before concluding that zero-click search took your traffic, read your own title and description against the query text and ask whether you would click.
It is an unglamorous check and it costs ten minutes. The appeal of the zero-click explanation is partly that it is external — it makes the problem an industry condition rather than an editorial mistake. That appeal is exactly the reason to test the boring hypothesis first. And when the generated answer genuinely is the competition, what gets cited is decided by different machinery than what ranks, which is a separate problem needing a separate fix.
We will publish the numbers again once there is enough data to say whether any of this worked.
Frequently asked questions
Why would a page rank in the top ten and get no clicks at all?
The most common cause is not the one people reach for. Before blaming zero-click search, check what your page actually displays in the results for that query. A title written for a different audience than the one searching will suppress clicks completely, even from position three, because the searcher reads the title, decides the page is about something else, and moves on.
How do you tell AI Overview absorption from a title problem?
You largely cannot, and that is worth stating plainly. Search Console does not report whether an impression appeared alongside an AI Overview, so the two hypotheses produce identical data. What you can do is eliminate the cheaper explanation first: read your own title and meta description against the query text and ask whether you would click.
What is an intent mismatch in a search result?
It is when the page genuinely covers what the searcher wants but the title describes something else. A page listing enterprise automation tools that is titled around reclaiming your morning will rank for tool queries on the strength of its body content, then fail to convert the impression, because the only thing the searcher sees is the title.
Does rewriting a title risk losing the ranking?
It can, and the risk is real enough to take seriously. Titles are a ranking input as well as a click input, so a rewrite that strips the terms the page ranks for can cost position. The safer move is to keep the terms the query data shows are working and change the framing around them, rather than replacing the title wholesale.
Why do auto-generated meta descriptions hurt click-through?
Because most generators truncate the opening paragraph at a character limit, producing a fragment that stops mid-sentence. The searcher sees text that breaks off without resolving, which reads as a low-quality page before they have visited it. It is a silent problem, since nothing in an audit flags a description that is the right length but ends mid-clause.
How long does it take to see whether a title change worked?
Longer than feels comfortable. Google has to recrawl the page, then accumulate enough impressions for a click-through rate to mean anything. For a page earning a few hundred impressions a month, a fortnight is not a result, and treating early movement as signal is how teams end up reverting changes that were working.
Related reading
How AI Overviews Choose What to Cite
Ranking and citation are different selections made by different systems. Why a page can sit in the top ten for a query and never once be quoted.
Generative AI in SEO: A Practical Guide to Optimizing for Google's AI-Powered Search
How generative search changes optimisation: what GEO means in practice, why E-E-A-T carries more weight, and where technical SEO still wins.
Why AI Security Frameworks Fail in Practice
Frameworks describe controls for a system with a boundary. LLM applications dissolve that boundary, and the checklist stays green while the gap opens.