Somebody has probably told you that AI is about to end search traffic, and you cannot tell whether that is true or whether it is a pitch aimed at you. So here is the most honest number available. On Google searches where an AI summary appeared, people clicked a traditional search result in 8% of visits, against roughly 15% on searches without one. That comes from the Pew Research Center, which watched the real browsing activity of 900 US adults in March 2025 instead of asking them what they thought they did. That is a genuine halving, and nobody selling you anything should be waving it away. But the same study found something almost nobody quotes, and it matters more to your business than the click number does: browsing sessions ended on 26% of pages carrying an AI summary, against 16% of pages without one. The AI answer is not only taking the click. Sometimes it is ending the search.
Key Takeaways
When an AI summary appears, clicks on a traditional result fall to 8% of visits from roughly 15%, and links inside the summary itself are clicked in only 1%. Being named in the answer is worth having, but on its own it is not a traffic plan.
A study of everyone's searches says nothing about whether the queries that bring you customers are the ones losing clicks. Your own Search Console can, and the check takes about twenty minutes.
That control does not touch Google Search. Only blocking Googlebot does, and that removes you from ordinary search results at the same time, which is why it is almost never the right lever.
Is this happening to your business, or just to the average?
Almost every article on this subject leads with a figure about all searches everywhere. The most-cited one comes from SparkToro's 2024 clickstream study, which found that 58.5% of US Google searches ended without a click to the open web. It is serious work, built on panels of real devices rather than estimates, and the same team later analysed 332 million queries over 21 months to look at how people actually use Google.
It still cannot tell you anything about your business, and it is worth being clear about why. That figure counts every search anybody ran: weather, sports scores, currency conversions, celebrity ages, and the several billion searches a year where nobody was ever going to visit a website. A plumber in a town of forty thousand and a national software brand are both inside that one percentage. Reading your own exposure off it is like reading your own rainfall off a national average.
There is a second limit that gets skipped. It is vendor-commissioned clickstream analysis, not a census from Google, and it counts a click to another Google property differently from no click at all. That does not make it wrong. It makes it a measure of direction rather than a measure of you.
The question worth twenty minutes is much narrower. Are the searches that actually bring you work losing their clicks? That has an answer, it is free, and it is sitting in your own account.
What actually happens when an AI summary appears?
The Pew study is worth understanding properly, because it is one of the few measurements of behaviour rather than opinion. Researchers recorded what 900 US adults did in their browsers during March 2025. Of those, 58% ran at least one Google search that produced an AI summary, so this is not a fringe experience affecting a handful of unusual queries.
Three findings carry most of the weight, and they point in slightly different directions.
| What was measured | With an AI summary | Without one |
|---|---|---|
| Clicked a traditional search result | 8% of visits | about 15% of visits |
| Clicked a source cited inside the AI summary | 1% of visits | not applicable |
| Browsing session ended on that page | 26% of pages | 16% of pages |
The click loss is the headline, and it is real. The second row is the one that should temper anyone's enthusiasm for citation as a traffic strategy: being named inside the answer earned a click in one visit out of a hundred. That is closer to being mentioned on the radio than to being sent a customer. It is still worth having, for reasons covered further down, but it is not a substitute for the click.
The third row is the finding that changes how you should think about the whole thing. When a session ends, the searcher did not go to a competitor. They got what they came for and stopped. That means the loss is not being redistributed to the sites that out-optimised you. For a certain kind of page, the demand is simply being served upstream.
Two honest limits sit beside those numbers, and anyone quoting them at you should say so. This was a panel of US adults over one month in early 2025, so it is a snapshot rather than a permanent state, and AI answers have kept changing since. And it measures what happens once an AI summary appears. It does not tell you how often one appears for the queries that matter to you, which is the part that decides your exposure.
Why does Google's own account say the opposite?
Google has published a different reading. In August 2025 it said AI in Search is driving more queries and higher-quality clicks, where a higher-quality click means a visitor who does not bounce straight back to the results.
You should know that claim exists, and you should also know what it is made of. Google defines the quality measure itself and has published no dataset anybody outside the company can audit. That is not an accusation of dishonesty; it is a statement about what kind of evidence it is. A measured panel study and a company statement about its own product are not the same class of thing, and where they disagree, you are entitled to weight them differently.
There is a version of this where both are true at once, and it is probably the closest thing to reality. Fewer people click, and the ones who do click are further along, because the AI answer already handled the part that never had commercial value to you anyway. If that is what is happening, the businesses hurt worst are the ones whose traffic was mostly the part that was never going to buy. Which is a very different problem from the one the headlines describe, and a much more fixable one.
How do you check your own exposure in twenty minutes?
Nothing in this section needs a tool you do not already have, and none of it needs us. All of it happens inside Google Search Console, free, on your own data. If your site is not verified there yet, that is the first job and it takes ten minutes.
The point of the exercise is to replace a market-wide percentage with your own, and to find out whether what you are feeling is an AI problem at all. It very often is not. Google publishes its own guide to debugging a traffic drop, and the shape of the drop is what tells you which cause you are looking at.
Four checks, about twenty minutes, no tools you do not already have
- 1
One, put impressions and clicks on the same chart
Open the Performance report and set the range to the last sixteen months. You are looking for one specific shape: impressions holding steady or climbing while clicks fall away underneath them. That widening gap is the signature of pages still being shown and no longer being visited.
- 2
Two, rule out the boring explanations first
If impressions fell too, this is not a zero-click story. That is a ranking, indexing or seasonality question, and treating it as an AI problem sends you fixing the wrong thing for a quarter.
- 3
Three, read your top thirty queries out loud
Sort by impressions. Count how many are questions with a single factual answer, and how many are somebody trying to hire, buy, book or compare. That ratio is your real exposure, and it beats any industry figure.
- 4
Four, split the same report by page
Filter to your top landing pages and run the same impressions-versus-clicks comparison on each. Losses will concentrate in a specific kind of page rather than spreading evenly, and knowing which kind is the whole point of doing this.
Write the numbers down with the date beside them. One reading is noise. The value of this check comes from repeating it in three months and having something to compare against.
Which of your pages are actually at risk?
Sort your pages by what the visitor wanted, not by how much traffic they get. Traffic tells you what you are about to lose. Intent tells you whether it mattered.
Pages that answer a closed factual question are the most exposed. What the standard height of a countertop is, how long a particular paint takes to cure, what an acronym in your trade stands for. If that is the whole page, an answer engine can now serve it completely, and it will. This is the category the alarming headlines are actually describing.
Pages answering a question with a general shape and a personal answer hold up far better. What this costs for a house like mine, how long my project will take, what happens if it rains halfway through the work. A summary can give the average. It cannot give yours, and readers know the difference well enough to keep looking.
Pages that exist to start a transaction are the safest of all, and this is the part that gets lost in the noise. Nobody books a job, requests a quote or hands over a deposit inside an AI answer. Those pages need to be fast, findable and easy to act on, which is a web design question more than a content one, but they are not the thing being taken away.
If your site is mostly the first category, you have real work ahead of you and it is better to know now. If it is mostly the second and third, your position is a great deal stronger than the headlines suggest, and the correct response is maintenance rather than panic.
Should you block the AI crawlers?
This question comes up constantly, and the honest answer is that it is a trade rather than a fix. Before making the trade, it is worth knowing that the most commonly recommended control does not do what most people think it does.
Google's crawler documentation separates these agents deliberately. Googlebot handles crawling for Google Search, and that includes the AI features inside Search. Google-Extended is a separate token covering Gemini training and certain Gemini grounding, and it has no effect on Google Search at all. So blocking Google-Extended will not take you out of AI Overviews. The only thing that removes you from those is blocking Googlebot, which removes you from ordinary search results at the same time. People get this wrong in a robots file and cause damage far larger than the problem they set out to solve.
OpenAI draws a similar line and documents it publicly. GPTBot governs whether your content may be used for training. OAI-SearchBot governs whether ChatGPT can read your pages to summarise and cite them, and OpenAI states plainly that content will not appear in ChatGPT summaries and snippets if you block it. Those are two different decisions and they deserve two different answers.
Now the actual trade. Blocking one of these agents can keep your content out of some AI training and some AI answers. It also removes you from the surfaces where buyers are increasingly asking who to hire. For a business whose content is the product, a publisher, a course seller, a research firm, blocking can be entirely rational, because the content being read without a visit is the loss. For a local service business trying to be recommended to someone who is ready to buy, it usually is not, because the summary is doing the introducing.
One more thing worth knowing before you write any rules. Not every agent respects them. Perplexity documents a user-triggered fetcher that visits a page because a person asked for it, and states that it generally ignores robots.txt. So access design and authentication still matter for anything you genuinely need kept off these surfaces. A robots file is a request, not a lock.
What should you fix this quarter?
None of this needs a rebuild or a retainer, and the order matters more than the list. These are ranked by what they return for the effort, not by what is most interesting to talk about.
Get the boring technical things right first, because they gate everything else. Your key pages should return a normal status, not be blocked from indexing, be reachable through ordinary links from your home page, and show their important information as selectable text rather than as pictures of text. Google states that a page has to be indexed and eligible to show a snippet before it can be used in AI features at all. Every visibility conversation that skips this step is decoration.
Answer the question properly, then give a reason to come anyway. The pages that survive this shift are not the ones that withhold information to force a click. Those get skipped, and they have earned it. The ones that hold up answer the question in the first sentence and then carry something a summary cannot: your actual prices, your actual process, photographs of your actual work, a form that books a real appointment. An answer engine can summarise a definition. It cannot summarise a quote for somebody's specific job.
Make your pages easy to read and easy to quote. Clear headings that match real questions, direct answers underneath them, plain structured facts instead of adjectives. This is ordinary search engine optimization done properly, and Google has been explicit that there is no separate AI schema and no special optimisation for AI features. The same work that makes a page quotable to a machine makes it faster to read for a person in a hurry. If you want the longer version of how citation actually works, we have written separately about generative engine optimization.
Put the parts of your presence an AI answer cannot replace in order. For a local business that means your Google Business Profile, your reviews, and your name, address and phone number being consistent everywhere they appear. When somebody asks an assistant who to call nearby, those signals are most of what it has to work with. That is local SEO and it has become more load-bearing rather than less.
How do you measure something Google does not report?
You cannot measure this directly, and anybody selling you a dashboard that claims otherwise is selling you a guess with a chart on it. Google does not report whether an AI summary appeared for a given query. So everything below is a proxy, and it is far better to work with proxies you understand than with a number that merely looks precise.
The impressions-versus-clicks gap, tracked monthly. This is the same check from earlier, repeated. One reading is noise; a trend across three or four months is evidence you can act on.
Click-through rate split by query type. Separate your informational queries from your commercial ones and watch them apart. If informational click-through is falling while commercial click-through holds, that is the shift showing up in your own numbers instead of in somebody's study, and it tells you exactly which pages to work on.
Referrals from the assistants themselves. Segment your analytics for visits arriving from ChatGPT and the other assistants. The volumes are usually small and that is fine. What you are watching is the direction and what those visitors do when they land, because a small number of people who arrive already convinced is worth more than a large number who bounce.
Whether Bing will tell you. Microsoft has begun surfacing an AI Performance report in Bing Webmaster Tools. Bing is not where most of your customers search, but it is currently more forthcoming about AI-surface data than Google is, and free directional data beats no data.
Whether people arrive already knowing things. This one is not a metric and it is the most useful signal on the list. If enquiries start coming from people who already know your process and roughly what you charge, they read about you somewhere, possibly inside an answer they never clicked. Ask new customers how they found you. That answer has started changing, and it will show up in conversation months before it shows up in analytics.
Keep the success measure fixed on calls, forms and booked work rather than on citation counts. It is the only number in this whole subject that pays anybody.
How exposed is your site, really?
1. You open Search Console and see impressions climbing while clicks fall. What does that most likely mean?
2. You block Google-Extended in your robots file. What happens to your appearance in Google's AI Overviews?
3. Which of your pages is hardest for an AI answer to replace?
Pick an answer to begin.
Frequently Asked Questions About Zero-Click Search
Is zero-click search the same as AI search? No, and the difference matters when you are reading older data. Featured snippets, local packs, knowledge panels, calculators and sports scores were all resolving searches without a click long before AI Overviews existed. AI answers are the newest and fastest-growing form of it, not the whole of it.
Do AI Overviews always reduce clicks? No. Pew observed lower clicks in its March 2025 cohort of searches that produced AI summaries, but that is an average across a panel, not a prediction for a given query or a given business. Commercial and local searches behave differently from factual ones.
Can I opt out of AI Overviews without leaving Google Search? Not cleanly. Google offers preview controls such as nosnippet and the robots meta tags documented here, and they affect relevant Search surfaces together. Suppressing your snippet tends to cost you normal visibility at the same time, which is why very few businesses find it a good trade.
How do I know whether ChatGPT is citing me? Allow OAI-SearchBot, then check your analytics for referrals from ChatGPT and test the questions your customers actually ask, by hand, a few times. There is no report that will tell you this, and any tool claiming a definitive citation count is estimating.
Do I need special AI schema or an AI-specific plugin? No. Google states there are no additional requirements or special optimisations for its AI features, and that the work is the same people-first work as ever. Anyone selling you an AI schema product is selling you something Google has said in writing you do not need.
Does my Google Business Profile still matter with AI answers around? More than it did, for a local business. It is a structured, verified statement of who you are, where you work and what people think of you, which is exactly the sort of thing an assistant can use with confidence. We have written on whether you still need SEO alongside a Business Profile, and the short answer has not changed.
Should I stop publishing content, then? No, but you should stop publishing the kind that is now being served upstream. Pages that only restate a definition were always the weakest thing on any site. The pages worth writing are the ones carrying something only you have: your pricing, your process, your finished work, your view of the trade-offs.
The Bottom Line
The click loss is real, it is measured, and pretending otherwise would be dishonest. On searches where an AI summary appears, clicks on a traditional result nearly halve, links inside the summary are clicked in one visit out of a hundred, and a quarter of those sessions simply end. Anybody telling you this is overblown is arguing with a study rather than with you. What they are usually missing is that the loss lands unevenly, and where it lands is something you can check yourself this week.
The conclusion runs the opposite way from the panic. When a buyer never reaches a page of blue links, the businesses that still get found are the ones whose sites are fast, clearly structured and factual enough for an answer engine to quote, and whose prices, process and booking live somewhere no summary can reach. Your website does not matter less in that world. It becomes the thing the machine reads before deciding whether to mention your name at all. The work is not exotic and it is not new. It is the same work that has always made a site worth visiting, done to a standard that now has a second audience.
If you would like a hand working out which of your pages are exposed and which are already fine, that audit is a normal part of what we do at Web Leveling. We build fast, custom sites and the search and AI work around them, and every site we build is owned outright by the business that paid for it. Tell us what is bothering you about your web presence and we will come back to you within one business day, in language you can act on without a translator.
Terms
The words this subject hides behind
Tap a term to see what it means.
Zero-click search. Any search where the person gets what they needed on the results page and visits no website. Featured snippets, map packs, calculators and knowledge panels all do it, so this is much older than AI.
AI Overview. Google's AI-generated summary at the top of some results pages. It sits inside Google Search, which is why Googlebot governs it and Google-Extended does not.
Googlebot. The crawler for Google Search, AI features included. Blocking it removes you from ordinary search results, so it is almost never the right lever.
Google-Extended. A separate control covering Gemini training and certain Gemini grounding. It has no effect on Google Search, which is the single most commonly misunderstood point in this subject.
GPTBot and OAI-SearchBot. OpenAI's two agents. The first is about training on your content, the second about whether ChatGPT can read your pages in order to summarise and cite them. They are separate decisions.
Clickstream study. Research built from what a panel of real devices actually did, rather than from survey answers or keyword tools. Stronger than opinion data, still a sample of one population at one time.
Impressions. How often your page was shown in results. Watching impressions and clicks together, rather than clicks alone, is what tells you whether you are being skipped or simply not shown.