More impressions, fewer clicks: how to investigate zero-click search
Diagnose falling search CTR using query mix, position, devices, result layouts, and qualified inquiries before attributing the decline to AI answers.
Establish a comparable baseline
Start with two Search Console periods of equal length using the same search type and filters. If demand is seasonal, add a comparison with the equivalent period last year. Record clicks, impressions, click-through rate, and position. Keep the sitewide totals for context, but do not use them as the entire diagnosis: the underlying mix of queries may have changed.
Google defines CTR as clicks divided by impressions. It describes interaction with a search result, not the proportion of people who completed their task without visiting a website. Investigating zero-click search therefore requires additional evidence. First rule out position losses, a different query mix, changes to the snippet, and shifts between mobile and desktop audiences.
Separate questions with different intent
Create stable groups for branded queries, vendor selection, comparisons, how-to questions, and short definitions. Compare the same pages, countries, and devices within each group. A search for “what is CPA” should not be evaluated as though it were a request for a paid acquisition audit. The expected answer and business value of a visit differ.
Check whether a page gained substantial impressions for new, broad questions. That can lower aggregate CTR while existing query performance remains unchanged. Reporting may omit some low-volume queries, so a query table will not necessarily reconcile perfectly with the chart. State which rows and impressions your analysis covers rather than inventing an explanation for every missing observation.
Test whether the mix explains the percentage
Here is an illustrative example. In the first period, a site receives 1,000 branded impressions at 20% CTR and 9,000 informational impressions at 2% CTR. That produces 380 clicks and an overall CTR of 3.8%. Later, branded performance stays unchanged while informational impressions reach 19,000 at the same 2% CTR. Clicks rise to 580, but overall CTR drops to 2.9%.
Nothing became less clickable within either group. When total clicks also decline, the same decomposition helps identify the group that actually deteriorated. Do not assign every impression without a click to an AI answer. Those observations can include ordinary comparisons, irrelevant appearances, abandoned searches, and many other behaviors that this report cannot distinguish.
| Query group | Earlier: impressions / clicks | Later: impressions / clicks |
|---|---|---|
| Branded | 1,000 / 200 | 1,000 / 200 |
| Informational | 9,000 / 180 | 19,000 / 380 |
| Total | 10,000 / 380 | 20,000 / 580 |
| Aggregate CTR | 3.8% | 2.9% |
Inspect the results for the affected questions
Select queries where click behavior changed under reasonably comparable conditions. Record the date, country, device, exact wording, and visible result features: AI answers, snippets, maps, videos, or ads. Preserve a screenshot and the observation notes. A manual search remains a sample; your own results do not reproduce every impression recorded for customers across the US or Canada.
Google includes its AI search features in overall Search Console reporting. Finding an AI answer beside a falling CTR is therefore a hypothesis worth investigating, not a causal estimate. Also check whether the page title changed, the page became less relevant to the question, or another page on your site started appearing for the same intent.
Compare business outcomes without forcing attribution
Review qualified inquiries and sales associated with the affected landing pages using your agreed analytics definitions. Separate inquiry quantity from quality. Ten relevant business conversations may matter more than fifty unsuitable submissions. For a longer sales cycle, compare cohorts that have had the same opportunity to mature; otherwise the newest period will look artificially weak.
Branded searches, direct visits, and self-reported discovery answers provide additional context. They do not prove that a person saw a particular AI response. Keep these signals separate and preserve their definitions. If form tracking failed, repair measurement first. Editing an article cannot reconstruct missing conversion events from the preceding weeks.
Choose a response that matches the evidence
If a commercial page lost position, examine intent fit and competing offers. If a short answer now fits within the results page, consider what a visit can add: a calculation using the reader’s inputs, a comparison of constraints, or a detailed worked example. Improve the specific page where that continuation is useful rather than rewriting the entire blog after one unfavorable chart.
Frequent mistakes include labeling every decline zero-click, combining countries with different demand, and crediting an article for every increase in brand searches. Record the decision, baseline, and review date. Evaluate the same query group after the change. When the evidence is insufficient, the next task is to collect better observations rather than rush a headline rewrite.
Put it into practice
Frequently asked questions
Is the zero-click rate equal to 100% minus CTR?
No. Impressions without clicks include many behaviors and do not establish that the person received an answer. Search Console does not record the user’s complete journey after an impression.
Should we remove articles that now get fewer clicks?
Not from that metric alone. Review intent, position, qualified outcomes, and internal navigation value. Consolidating or removing a page requires a separate assessment of its usefulness and overlap.