A zero-click search is a search that ends inside Google: the user finds the answer on the results page and does not open any website. The query is solved, the site that held the information never receives an HTTP request, and in its analytics that person never existed.
The phenomenon is almost always reported as a single percentage, usually an American one. That is the quickest way to make the wrong decision. On two baskets of Italian keywords we track, measured on the same day on the same engine, the share of queries with an AI-generated answer on the page is 8.1% in the first and 88.1% in the second.
Seven surfaces can close a search, not one
Since 2024 zero-click has been equated with AI Overviews. They are the newest element, not the most frequent. A SERP can close a search in at least seven different ways, and five of them existed long before generative AI.
- The AI-generated summary at the top of the page.
- The featured snippet, which lifts a paragraph from a single page and cites the source.
- The People Also Ask box, which opens four more answers without leaving the SERP.
- The knowledge panel, which answers entity queries: who it is, where it is, when it was founded.
- The local pack, where calls and directions are conversions that happen without anyone visiting a website.
- Image packs, product carousels and review carousels, which satisfy the visual and reputational side of the question.
- Direct widgets: weather, calculator, conversions, opening hours, scores.
The distinction matters because each of these surfaces is won and lost in a different way, and because the mix changes radically from one sector to another, as the SEO statistics we have collected show. Anyone treating zero-click as an AI problem is optimising for the wrong surface.
The figures in circulation do not measure the same thing
The percentages you read around, from 58% to 77%, look contradictory. They are not: they measure different things, with different methods, in different markets. Before quoting one, it is worth knowing which.
| Source | What it measures | Value | Market and sample | Period |
|---|---|---|---|---|
| SparkToro, Datos clickstream data | Google searches that produce no click at all | 58.5% US, 59.7% EU | United States and European Union | Sep 2022 - May 2024 |
| SparkToro, Datos clickstream data | Clicks to the open web per 1,000 searches | 374 US, 360 EU | United States and European Union | Sep 2022 - May 2024 |
| Pew Research Center | Clicks on a traditional result, with and without an AI summary | 8% with, 15% without | 900 US adults, 68,879 searches | March 2025 |
| Pew Research Center | Clicks on a source cited inside the AI summary | 1% | same sample | March 2025 |
| Pew Research Center | Sessions ended straight after the SERP | 26% with AI, 16% without | same sample | March 2025 |
| Bain & Company with Dynata | Consumers who rely on zero-click answers for at least 40% of their searches | 80% | survey, n=1,117 | December 2024 |
| Semrush | Keywords that trigger an AI Overview | 6.49% in January, 24.61% in July, 15.69% in November | 10 million keywords | 2025 |
| SEOZoom Observatory | Italian keywords that trigger an AI Overview | 77% | Italy, over 18 million keywords | June 2026 |
The SparkToro figure counts how many searches end without a click, including those where the user clicks on a Google property such as YouTube or Maps: that is why Rand Fishkin presents it in its most readable form, clicks reaching the open web per thousand searches. The Pew figure measures something else again: the real behaviour of 900 people whose browsing was tracked, and it is the only one that isolates the effect of the AI summary on the same users. The Semrush and SEOZoom figures are not about clicks: they count how many queries show an AI Overview, which is a condition, not an outcome.
The most repeated estimate of all, an organic traffic drop of between 15% and 25%, comes from the Bain study run with Dynata on 1,117 people in December 2024. Bain presents it as an estimate, not a measurement. It is quoted everywhere as if it were a measured number.
Penetration depends on the vertical
The clearest sector breakdown we have found comes from Italy, where the SEOZoom Observatory measured AI Overviews in June 2026: of more than 18 million keywords monitored on google.it, 77% trigger one. On the dedicated sector samples the gap is wide: Health at 51.74%, Finance at 33.8%, Law and government at 28.92%, Business and industry at 20.54%, Travel and tourism at 11.03%, Restaurants and nightlife at 9.88%. The absolute levels belong to the Italian market; the spread between sectors is the point, and it is the same pattern we see when we check UK SERPs.
Two baskets, 8% and 88%
National averages are good for headlines. To decide where to invest, you need your own number. We took two Italian projects we manage, 278 keywords in total, both tracked on google.it in Italian, and counted how many queries show each surface. Data collected on 15 September 2026.
- Basket A: 135 keywords for a B2B manufacturer, mostly product and price queries.
- Basket B: 143 keywords for a solar energy company, queries about leasing and selling land, so with legal and financial content.
| SERP surface | Basket A: B2B manufacturing (135 keywords) | Basket B: solar (143 keywords) |
|---|---|---|
| People Also Ask | 94.8% | 100% |
| AI-generated answer | 8.1% | 88.1% |
| Image pack | 77.8% | 1.4% |
| Reviews | 69.6% | 1.4% |
| Local pack | 24.4% | 0.7% |
| Knowledge panel | 3.0% | 9.8% |
| Paid ads | 0.7% | 14.7% |
Two companies, the same engine, the same day, two scenarios with almost nothing in common. In the manufacturing basket the AI answer is a rare event and the real threat lies elsewhere: an image pack on 77.8% of queries and reviews on 69.6% push the organic results below the fold long before any generative model steps in. In the solar basket the opposite happens, with the AI answer on almost nine queries out of ten and the SERP behaving like the Finance and Law samples measured by SEOZoom, only much higher.
The same split shows up on Google UK. We ran the English equivalent of each basket's head query (DataForSEO, Google UK, desktop, September 2026). "solar farm land lease" opens with an AI Overview, followed by a People Also Ask box whose four questions (how much farmers are paid per acre, the minimum acreage for a solar farm and so on) are each answered by AI. "led video wall price" has no AI Overview at all: the organic results are interleaved with three product carousels, 30 priced product cards in total, plus a People Also Ask box. Two B2B searches, two completely different ways of keeping the user on the page.
The only constant is the People Also Ask box: present on 94.8% of the first basket and 100% of the second. It is the most widespread zero-click surface in the SERPs we track, and hardly anyone treats it as one.
First place, one click per hundred impressions
The presence of a surface does not tell you what it costs. For that you need to cross positions with Search Console data. For basket A we took the 100 keywords with the most impressions out of the 135 tracked and added up their impressions and clicks.
- Total: 6,336 impressions, 63 clicks. CTR 0.99%.
- The 29 keywords in positions 1 to 3: 4,128 impressions, 50 clicks. CTR 1.21%.
- The 10 keywords in first position: 1,139 impressions, 12 clicks. CTR 1.05%.
- The 20 keywords beyond position 10: 561 impressions, zero clicks.
The most telling single case is the sector's head query, an estimated 8,100 monthly searches in Italy: a stable position 1, 953 impressions, 11 clicks, a CTR of 1.15%. The variant written as one word, in second position, brings 2,129 impressions and 27 clicks, 1.27%.
Two caveats, so the number is read for what it is. The tracked position is the desktop one on google.it, while Search Console aggregates desktop and mobile, and on mobile the same SERP scrolls much further before reaching the organic results. And an impression is counted even when the result never entered the user's field of view. Both push CTR down, and both are exactly the mechanism we are talking about: the site was there, the person did not see it or did not need it.
It is worth holding both facts together. On these same keywords the project has an excellent average position, built over the journey described in our manufacturing SEO case study: from position 88 to position 2, with a 25.55% share of voice, ahead of Amazon. Positions are holding. Clicks per impression are not. The two have stopped moving together, and reporting only the first is why many SEO reports look great while the sales team receives no enquiries.
Which queries hold up and which do not
Not all lost queries are worth the same, and not all were recoverable. Sort them by search intent, which is the best predictor of how exposed you are.
- Definitions and one-line questions. What it is, how long it lasts, how it is calculated. They are the first to close on the page and the hardest to defend. The Pew figure puts it plainly: when an AI summary appears, only 1% of visits produce a click on one of the cited sources. Being cited does not bring traffic back; it builds recognition.
- Comparisons and lists. Best, alternatives, X vs Y. They hold up better because the choice requires seeing real prices, terms and availability.
- Transactional and price queries. In basket A, queries containing "price" and "prices" keep bringing clicks where the others do not. Someone who needs a quote opens the website.
- Brand queries. The most resilient, and the only metric that grows if zero-click is working in your favour.
The practical consequence is unpleasant but useful: on part of the informational queries the battle for the click is over, and persisting with content designed to win it is time badly spent. Better to move that time to the queries that still convert and to getting cited where you are not clicked.
How to measure your zero-click rate
The number you need is not in any study. You build it in five steps with the tools you already have.
- Separate brand queries from the rest. In Search Console, filter on the company name and its spelling variants, then look at the two groups separately. Keeping them together hides everything: brand queries have high CTRs and mask the collapse on the rest.
- Count the surfaces on your real basket. Take the keywords that actually matter to you and check how many show an AI answer, a featured snippet, a People Also Ask box, a local pack. This is the count that produced the table above, and any rank tracker that records SERP features can do it.
- Calculate CTR by position band, not as an average. Positions 1-3, 4-10, beyond 10. A single average across the whole site says nothing, because it mixes brand queries, service pages and articles.
- Compare with the same quarter of the previous year. Stable or growing impressions with falling clicks at the same position is the signature of the phenomenon. If impressions are falling too, the problem is something else and needs looking for elsewhere.
- Look at the Generative AI reports in Search Console. On 31 August 2026 Google extended them to all sites: they show your pages' impressions inside AI Overviews and AI Mode. We explained how to read them, and what their limits are, in the article on the impact of AI Overviews on SEO.
What to do when the answer stays on the page
The actions that change anything are few and dull.
- Answer the question in the first lines, in a form that can be extracted. It works for the featured snippet and it works for the models: the same structure serves both. It is the core of answer engine optimisation.
- Put in the text what the summary cannot contain: the real price, the delivery time, the configuration, the regulatory constraint, the number measured on your own case.
- Cover the People Also Ask questions with dedicated sections, since in the two baskets the box appears on between 94.8% and 100% of queries.
- Keep your structured data and business profile in order, because knowledge panels and local packs feed on them.
- Check how the brand is described by the models, not just whether it appears. We have collected the tactics in the guide to generative engine optimisation and in the one on how to appear in ChatGPT answers.
- Shift part of your targets from traffic to brand demand and assisted conversions.
If you want your own site's number instead of someone else's average, that is the work we do in our SEO and AI search visibility engagements, and the results on real cases are in our projects.
Zero-click searches: frequently asked questions
They are searches where the user gets the answer directly on the results page and does not open any website. The query is solved by an AI Overview, a featured snippet, a People Also Ask box, a knowledge panel, a local pack or a direct widget such as weather or the calculator.
It depends on what you measure. SparkToro, using Datos clickstream data collected between September 2022 and May 2024, reports 58.5% in the United States and 59.7% in the European Union, and puts it another way: for every 1,000 Google searches only 374 clicks reach the open web in the US and 360 in the EU. The Pew Research Center, on 68,879 searches by 900 US adults in March 2025, measured clicks on a traditional result in 8% of visits with an AI summary against 15% without.
No. On a basket of 135 B2B manufacturing keywords the AI answer appears on 8.1% of queries, while the image pack appears on 77.8% and reviews on 69.6%: those are the surfaces pushing organic results below the fold. On a second basket of 143 solar energy keywords the AI answer appears on 88.1% of queries. The mix changes from sector to sector.
The signal is a gap opening between impressions and clicks at the same position. In Search Console, compare the same quarter of the previous year with brand queries separated from the rest: if impressions are stable or growing and clicks are falling while the average position is not getting worse, traffic is stopping on the SERP. If impressions are falling too, the cause is something else.
Brand, transactional and price queries, because they need information the engine does not have or an action that has to happen on the website. Definitions and one-line questions are the most exposed: when an AI summary appears, according to the Pew Research Center only 1% of visits produce a click on one of the cited sources.
Yes, with a different unit of measurement. Rankings still decide which sources the engine reads and cites, so they remain the prerequisite for visibility inside generated answers. What changes is the reporting: alongside sessions and clicks you need to measure impressions, presence in SERP surfaces, citations in AI engines and brand search volume.
One thing the industry rarely says: the metric that replaces the click is not the citation inside the AI answer, it is brand search volume. The citation is decided by the engine and changes from one week to the next. People who, after reading an answer, search for your name instead of your product are the only sign that clickless visibility left something behind. It is a number that sits in Search Console, free, and almost nobody reports it to clients.
