Zero Click Search: How to Tell an AI Traffic Loss From a Core Update in Your Own Search Console

What a zero click search actually is

A zero click search is a search where the user gets their answer on the results page itself and never clicks through to a website. The query is satisfied by Google, not by you.

Here is how to tell whether that is what happened to your site, because the symptom looks identical to four other causes, and in nine years of doing this I have watched people spend entire quarters rewriting content to fix a problem they never had.

Three surface types produce a zero click result. The featured snippet, which lifts a passage out of a page and shows it at the top. The knowledge panel and the People Also Ask block, which answer factual and follow up questions inline. And the AI Overview, which synthesises an answer from several sources and links to them as supporting citations.

The term now covers two genuinely different behaviours, which is why the published numbers disagree so badly.

The old behaviour is featured snippet absorption, and it has existed since 2014. One page gets extracted, one answer gets shown, the query resolves without a click. Predictable, measurable, fought over.

The new behaviour is AI Overview and AI Mode absorption. Several sources get synthesised into one generated answer. No single page is being extracted, so no single page can reliably win the slot, and the citation links sit outside the answer body in a chip row or a sidebar. This is the shift that turned SEO into GEO, and it is why the 2019 zero click playbook does not transfer. I have set out the wider argument about what actually happened to clicks, impressions and rankings from 2024 to 2026 separately. This page does the narrower job: working out whether it happened to you.

Google’s own definition of impressions, position and clicks is blunt about how this lands in your reporting. An AI Overview occupies a single position in search results, and every link inside it is assigned that same position. Clicking an external link in an AI Overview counts as a click. Standard impression rules apply. So AI Overview activity is already inside your ordinary web search totals, blended in, not broken out.

That blending is the entire diagnostic problem.

Nobody searches for "zero click search"

One thing worth saying before the method, because it shapes how you should read the rest.

I went looking for how people actually describe this problem, in their own words, on the forums where they ask for help. Across thirty-three threads on BlackHatWorld, the Google Search Central community and the Shopify community, the phrase "zero click search" appears exactly nowhere.

What people actually write is this:

"Till about a month back, clicks data for our site followed the impressions data very closely. Then something happened, impressions are improving, but clicks are not. How do I go about figuring out what the issue is?"

That is from BlackHatWorld, and the date on it is 9 March 2022. Two years before AI Overviews existed. Hold onto that, because it matters later.

Or this, from February 2026:

"I’m seeing a noticeable organic clicks drop starting around mid of September 2025 and it’s not just my site, competitors in the same niche show very similar patterns. Would like to know how you diagnosed it and what actions did you take that actually helped."

The community has its own name for the chart shape, too. They call it the crocodile effect, because the impressions line and the clicks line open apart like a jaw. Somebody described it as "a jagged little graph that opens up like a jaw, sharp teeth and all", which is better writing than most of what gets published about it.

"Zero click search" is a term that lives in vendor blog titles. The people with the actual problem say "impressions are improving, but clicks are not", or "is it just me or everyone", or "I no longer receive traffic from Google". I have written this piece for those people, so the headings answer their questions rather than the industry’s.

Why your traffic dropped: five causes that look identical in a dashboard

Five separate things make clicks fall, and in a clicks chart all five look the same. A line goes down. The line does not tell you why.

The fix for each is different, and three of them are actively harmful if applied to the wrong cause. Rewriting good content because you assumed AI absorption, when you actually shipped a canonical bug, costs you the quarter and the bug is still there.

Here is what separates them.

The five signatures. Read it left to right, not top to bottom.

Read that table left to right, not top to bottom. The clicks column is nearly useless. The position and impressions columns do all the work.

The measurement artifact: impressions down, position better, clicks unchanged

Start here, because if this is your answer, nothing happened to your traffic at all and every other investigation is wasted.

On 12 September 2025 Google disabled the &num=100 parameter, which had let anyone request 100 search results on one page instead of ten. Rank tracking tools used it constantly, because one request got them a hundred positions. Every one of those requests registered in Search Console as impressions for every site on the page, down to position 100.

When Google switched it off, that inflation stopped. The independent consultant Brodie Clark documented what followed, and the numbers came from three separate datasets: 87.7 percent of Search Console properties saw impressions drop (LOCOMOTIVE, 319 sites), the median loss was around 15 percent (ThotSEO, 1,334 sites), and the aggregate decline in the following week was 25 percent (JetOctopus, 1,000 sites). Desktop was hit far harder than mobile, which is exactly what you would expect if rank trackers were the cause. Clark’s own summary: our Search Console data had been polluted with bot impressions for about a year.

Two consequences, and both matter more than the headline.

First, some of the impressions-up-clicks-down pattern that the entire industry attributed to AI Overviews was never human. Those impressions came from software. You never lost those clicks, because you never had those readers.

Second, and this one caught me: average position appeared to improve overnight. Positions 21 to 100 stopped being counted, so the average of what remained looked better. If you were reading position first, as I now tell everyone to, September 2025 would have told you a lovely story about improving rankings. Nothing had improved.

So the check runs first, and it is quick. Open the last 16 months, look at 12 September 2025, and ask whether your impression line steps down there, whether the drop is desktop-weighted, and whether clicks carried on unaffected. Impressions down with clicks flat is not a traffic loss. It is your data getting more honest.

The same logic covers the other artifacts. Google logs Search Console data anomalies on a public page, and it has logged plenty: Discover and generative AI logging errors starting 21 May 2026 and 24 June 2026, FAQ rich results disappearing from 7 May 2026, an impression reporting error running from 13 May 2025 to 27 April 2026. Note those first two dates. They are the exact start dates of the May 2026 core update and the June 2026 spam update. Anyone comparing before and after those dates would blame an algorithm for a logging bug, and nothing in the interface would warn them.

Impressions up, clicks down: the AI signature

If impressions held or grew, average position did not move, and clicks fell, that is AI answer absorption. Once the artifact above is ruled out, nothing else produces that combination.

The mechanism is simple once you see it. You are still being retrieved. You are still being shown. The user is being satisfied before they reach the decision to click. Position is unchanged because your ranking is unchanged.

The Pew Research Center measured this directly rather than modelling it, which is why I trust it more than most vendor studies. Tracking 900 US adults across 68,879 Google searches in March 2025, Pew found users clicked a traditional search result on 8 percent of visits where an AI summary appeared, against 15 percent where none did. They clicked a link inside the AI summary itself on 1 percent of visits. They ended the browsing session entirely on 26 percent of visits with an AI summary, against 16 percent without.

That last number is the one nobody quotes and the most important. The AI summary does not just take the click. It ends the session.

Impressions down, position down: the update signature

If impressions and average position both fell, you were re-ranked. That is a core update or a spam update, and the first move is to match your date against Google’s own record of confirmed updates rather than guessing.

This is where my worst diagnosis came from. An India based services client, clicks down, and I went straight to AI absorption because that was the story everyone was telling that year. It was a core update. The position column had moved and I had not looked at it, because I went looking for the answer I expected to find.

The tell I missed: core updates roll over days or weeks, not overnight. The March 2026 core update ran 27 March to 8 April, twelve days. The May 2026 core update ran 21 May to 2 June. A twelve day re-ranking does not look like a cliff. It looks like drift, and drift is exactly what gradual AI absorption is supposed to look like. That is why the two get confused.

Look at the position column on that account before you file it, though, because it bends the rule above. From 1 June to 25 August 2025 it took 474 clicks and 51,400 impressions at an average position of 19.4. From 1 September to 20 October 2025 it took 20 clicks and 2,300 impressions, and average position improved to 15. When a demotion is deep enough, pages drop out of the results entirely rather than sliding down them, and the queries left behind are the few still ranking well. The average improves because the losers left the sample. So the update signature is impressions and position falling together, unless the fall is deep enough to empty the results, in which case impressions collapse and position looks healthy. The date and the impressions column give it away, not the position.

Both down on one date: the technical signature

If impressions and clicks both fall off a cliff on a single date, and especially if the fall is concentrated on one page template, it is technical. A noindex shipped with a release. A canonical pointing somewhere wrong. A redirect chain. A rendering change that stopped the content resolving for Googlebot. The mechanism matters here: Googlebot has to fetch, render and resolve your content before any of it can rank, and any break in that chain removes the page from the index rather than demoting it, which is why the line falls vertically instead of drifting.

Algorithms do not work in clean vertical lines. Deployments do. If the chart has a corner in it, look at your deployment log before anything else. One of the Search Central threads is titled "92% Decline in Search Impressions and Clicks in 2 Days", and the two days in that title is the diagnosis, sitting right there in the subject line.

Position down, competitor up: the boring answer nobody checks

If position fell on a defined set of queries while impressions held roughly steady sitewide, somebody out-ranked you. That is all that happened.

This is the least interesting answer and the one clients resist hardest, because "we were beaten" is a harder conversation than "the algorithm changed". It is also the easiest to verify: take the ten queries that lost the most clicks, search them, and look at who is above you now who was not there before.

The cause has changed by phase, and so must the diagnosis

Almost every article on this topic writes as though there is one answer to "what is taking my clicks". There is not. The dominant cause has shifted in phases, and a diagnosis that ignores the date of your drop is guessing.

Confirmed Google update windows from the Search Status Dashboard, set against the AI rollout.

Two things fall out of that which are worth more than any single statistic.

The first is that "SEO traffic drop" in 2024 and "SEO traffic drop" in 2026 are different diagnoses with different base rates. If your drop was in late 2024 and you concluded AI, you were probably wrong. If your drop was in early 2026 on informational queries, you were probably right.

The second is that the phases track AI search maturing into GEO. In phase one the question was "will an AI answer appear". By phase four it is "am I the source it cites". Those need different instruments, and Search Console only partly provides the second.

There is a third thing, and it is uncomfortable. On the forums, when the June 2025 drops hit, the community consensus was that weak backlink profiles and thin content took the damage. The vendor content published that same fortnight said AI Overviews did it. Two different causal stories about the same two weeks, and the one that sold more software won the search results.

What that evolution looks like in four real brands

Phase tables are easy to write and hard to trust. So here is the same period read through four live domains, chosen because they sit at completely different scales and in different verticals.

One caveat before the numbers, and it is not a formality. These are Ahrefs estimates of organic traffic, not measured traffic. Measured traffic comes from Google Analytics and Search Console. I use Ahrefs here because it is the only way to see a decade of history for a domain you do not own. Ahrefs has also just made its adaptive traffic model the default, which reweights SERP features including AI Overviews and refreshes monthly, so part of any year on year movement below is model change rather than site change. Treat the direction as signal and the precise figure as noise.

Estimated organic traffic, worldwide, sampled on the same date in early May each year. Ahrefs, read 19 September 2026. Domains withheld.

Three things fall out, and the third is the one I would put in front of a client.

The portal’s collapse has nothing to do with AI. It lost roughly half its estimated organic traffic between 2016, when it sat at 9.69 million, and 2024. Every one of those years predates AI Overviews. And through the AI rollout, 2024 to 2026, its estimated traffic went up. If you ran that site and blamed AI for your decline, you would be blaming a nine year trend on a two year event.

The accountancy practice grew straight through it. A young site, indexed only from 2026, climbing all year. AI Overviews did not stop a new site in a commercial vertical from gaining.

The two brands that did fall lost keywords far faster than they lost traffic. This is the finding worth the section.

Traffic down a quarter to a half. Ranked keyword count down three quarters to four fifths. Those two numbers moving at such different rates tells you what happened: the head terms held and the long tail disappeared.

That is precisely the shape AI absorption should produce. Semrush found 82 percent of desktop AI Overviews in its 200,000 keyword study appeared on keywords with under 1,000 monthly searches. The long tail is where AI answers live, and the long tail is what these two sites lost.

It is also the shape a keyword database refresh produces, which is why I would never call this diagnosed from Ahrefs alone. It is a hypothesis. You confirm it in Search Console, on the account, by checking whether the lost queries were informational long tail and whether position held on the head terms. The tool suggests. Only the account decides.

The GEO layer, which Search Console will not show you at all

The same four domains, read against Ahrefs’ AI citation index on 19 September 2026. This is the number of AI responses across platforms in which each domain appears.

AI responses containing each domain. Ahrefs AI index, all locations, 19 September 2026.

Look at the bottom two rows. Both appear in Google’s AI surfaces at reasonable volume for their size. Both are close to invisible in ChatGPT and Copilot. The accountancy practice appears in one Copilot response. The developer appears in none.

That is a different problem from the one this article diagnoses, and the one Search Console structurally cannot see. Google’s generative AI report covers Google’s own AI surfaces. It has nothing to say about whether ChatGPT or Copilot has ever heard of you. If your visibility problem is that row of near zeroes, no amount of Search Console analysis will find it, because the data is not in there.

The Search Console decision tree

Everything below uses free Google Search Console data. No paid tool is required, and I would rather you did this in Search Console first anyway, because third party tools model traffic and Search Console counts it.

1. Rule out the measurement artifact. Check 12 September 2025 on a 16 month view. Check Google’s data anomalies page for your drop date. Check whether someone added a new property. If impressions fell while clicks held, stop, you are fine.
2. Rule out Google. Open the Search Status Dashboard and check whether your date sits inside a confirmed core or spam update window. Two minutes, and it resolves a meaningful share of cases.
3. Set the comparison window. Last 16 months, compare, period over period and then year over year.
4. Read the position column first. Before clicks. Before impressions.
5. Split the query set by intent. Informational, navigational, commercial, separately.
6. Isolate the pages. Concentrated on one template is technical. Concentrated on informational content is AI. Spread evenly is an update.
7. Check the generative AI performance report, if you have it. Worldwide since 31 August 2026. Impressions only, no clicks.

Setting the comparison window so seasonality does not fool you

Use the last 16 months and compare like for like. Google’s own traffic drop debugging guide tells you to compare the last three months to the previous period and then the last three months year over year, and it says that for a reason: the two comparisons answer different questions.

Period over period tells you what changed recently. Year over year tells you whether what changed is seasonal. If you only run one, run year over year, because seasonality is the most common false positive I see, and the December quarter is a trap for anyone comparing to November.

Three mechanical things will mislead you if you do not know them.

• All Search Console dates are Pacific Time. If you are in India or the UK, your day and Google’s day do not align, and a one day event smears across two rows.
• The newest data is preliminary and can still change. Never diagnose a drop using the last three days. It always looks like a cliff and it usually is not.
• A domain property and a URL prefix property for the same site legitimately report different totals, because the domain property includes every subdomain and protocol. If your numbers changed when someone added a property, nothing happened to your traffic.

Splitting informational from commercial queries

This is where the diagnosis is actually made, and Search Console will not do it for you.

Export your queries and split them by intent. Question words first: what, why, how, when, who, is, does, can. Those are informational and where AI absorption lands. Then brand terms, which are navigational and should be almost immune. Then everything with a buying modifier: price, cost, best, near me, buy, hire, services, review.

Now compare CTR change across the three groups separately.

If informational CTR fell hard and commercial CTR held, that is AI absorption, and it is a content strategy problem, not a ranking problem. If all three fell together, it is not AI. AI does not touch navigational queries. A drop that includes your brand terms is an update, a tracking break, or a genuine demand collapse.

One caveat that matters more than it sounds. The moment you apply any query filter, anonymised queries silently drop out of the total, because Google excludes low volume queries to protect user privacy and only includes them in unfiltered chart totals. Your filtered number will be lower than your unfiltered number, and the gap is not lost traffic. I will come back to how large that gap gets.

Reading the position and impression columns together

This is the check I run first now, and two years ago I did not run it at all.

Tick average position and impressions before you look at clicks. Clicks are the symptom. Position and impressions are the diagnosis.

There is a trap inside average position and it catches people constantly. Google defines average position as the topmost position your property occupied, averaged across all the queries you appeared for. It is not where you rank. It moves when your query mix moves, even if not a single ranking changed. Pick up a thousand impressions on a weak new query set and your average position gets worse while your rankings are untouched. Lose a hundred thousand bot impressions from positions 21 to 100 and it gets better while nothing improved.

My own property demonstrates this uncomfortably well. Over the last 365 days itzfizz.com took 646,800 impressions and 6,200 clicks, a click through rate of 1.0 percent. Read in isolation that looks like textbook AI absorption. It is not. The average position over that window is 37.8. Nobody clicks position 38. The low CTR is explained entirely by ranking depth, and if I had blamed AI Overviews for it I would have been treating the wrong disease.

That is the whole point of reading position first.

The opposite trap shows up on accounts that are growing. One client account I work on grew its clicks over sixteen months, and its click through rate still settled at 0.3 percent at an average position of 19.5, because impressions grew faster than clicks did. If you only watched the impressions line you would report a good year. If you only watched the click through rate you would report a disaster. Neither reading is right, and you only get the truth by reading the two columns against each other.

Which query types actually lost clicks, and which never did

Informational queries took almost all the damage. Transactional and navigational largely did not.

Advanced Web Ranking measured AI Overview prevalence across 8,000 keywords in July 2024 and found it split sharply by intent: informational 21.29 percent, commercial 3.49 percent, transactional 0.47 percent, navigational 0.09 percent. Semrush’s 200,000 keyword study, on September 2024 data, found 80 percent of desktop keywords triggering an AI Overview were informational.

Those figures are dated, and the direction of travel since is the important part. Semrush re-ran the question across more than 600,000 keywords for the six months from November 2025 to April 2026 and found AI Overview appearance on commercial intent SERPs grew 71 percent, while transactional intent fell 5 percent. Finance grew 231 percent. BrightEdge put AI Overview presence on tracked queries at roughly 48 percent in February 2026 against roughly 30 percent a year earlier.

So the honest position for 2026: informational queries were hit first and hardest, commercial queries are now being absorbed too and fast, and your vertical changes the answer by an order of magnitude. Health sat at 67.47 percent AI Overview prevalence in the 2024 study while ecommerce sat at roughly 1 percent. If you sell products, your transactional pages are largely intact. If you publish health, finance or education explainers, they are not.

You can see this on your own keyword set without guessing. Filter a keyword export by SERP feature and read which of your queries currently carry an AI Overview, then cross-reference those against the queries that lost clicks.

Two caveats before you use any of this on a client. These are prevalence rates measured on sampled keyword sets, not on your keyword set. And prevalence is not damage: an AI Overview appearing on your query does not automatically mean you lost the click, which the next section makes uncomfortably clear.

The published numbers, and why they disagree

The headline zero click figures range from roughly 58 to roughly 68 percent depending on who you read, and the spread is not because anyone is lying. They are measuring different things.

SparkToro, working from Similarweb clickstream data, put US zero click searches at 68.01 percent for January to April 2026, against 60.45 percent for 2024. That is a panel measuring browser behaviour.

Seer Interactive, working from aggregated Search Console data across 53 brands and 5.47 million queries, measured organic CTR with and without an AI Overview present. February 2026: 2.36 percent with, 3.82 percent without. December 2025: 1.31 percent with, 3.97 percent without.

Ahrefs, comparing 150,000 keywords with AI Overviews against 150,000 without on December 2025 data, found AI Overview presence correlating with a 58 percent lower click through rate for the top ranking page, revised up from the 34.5 percent they published in April 2025.

And here is the finding that should make everyone slow down. In Seer’s January 2025 data, organic CTR was higher with an AI Overview present, 3.19 percent, than without it, 2.93 percent. The relationship inverted during 2025. It was not always negative. That single data point should stop anyone treating AI Overview presence as an automatic loss, and it gets stripped out of almost every retelling of the study.

All of these are correlations reported as correlations. None of them is your site. For the twelve month view across the accounts I actually work on, and what it means for the industry, I have written that up separately.

What to do for each of the five diagnoses

There is no single answer, and any article offering you one has not done the diagnosis.

If it is a measurement artifact. Nothing. Rebaseline your reporting from the artifact date forward, tell whoever reads your reports why the line stepped, and go back to work. The most valuable outcome of a diagnosis is often discovering there is nothing to fix.

If it is AI absorption. Stop optimising for the click on those queries and start optimising for the citation. Direct answer in the first two sentences of each section, self contained sections, comparisons in tables, question form headings. Then move commercial intent content up the priority list, because that is where clicks still exist. Accept that some informational traffic is not coming back, and rebuild the funnel around pages that still convert.

If it is a core update. Do nothing for at least two weeks after the update window closes. Rankings move during the rollout and acting mid-update means reacting to noise. Then audit the pages that actually lost position, one by one. Sitewide rewrites are almost always the wrong response.

If it is technical. Fix the bug. The good news: it is the one cause with a clean, fast, verifiable fix. Check URL Inspection on an affected page, compare rendered HTML against source, check canonicals, check robots.txt, check the deployment that shipped on the date of the cliff. If you want the technical layer swept quickly first, our free SEO checker covers the usual suspects.

If it is a competitor. Look at what they did. Usually it is more specific content, better first hand evidence, or a link position you do not have. Ordinary competitive SEO, and it responds to ordinary competitive SEO.

The point of a diagnostic is that the fix differs. If your answer to all five is "write better content", you have not diagnosed anything.

What this does not tell you

Search Console cannot show you an AI answer where no click occurred. That is the ceiling on this whole method, and you should know exactly where it sits.

Since the worldwide rollout completed on 31 August 2026, Search Console does include a generative AI performance report covering impressions from AI Overviews and AI Mode. Three limits matter. It reports impressions only, not clicks. It does not separate AI Overviews from AI Mode. And its data is also counted inside your ordinary web search totals, so it is a view, not a separate channel.

There is still no AI Overviews option in the search appearance dimension. Google enumerates 24 search appearance types and none is AI Overviews or AI Mode. Google also states plainly that clicks are assigned to a URL, not to a URL plus feature, so even where you can filter by appearance you cannot cleanly attribute clicks to it.

This is worth saying bluntly because a lot of vendor content implies the attribution problem is solved. It is not. On the Search Central forum someone asked, in as many words, whether there is any way to identify traffic coming from AI Mode or AI Overview. The honest answer is no. You can infer it by triangulation. You cannot isolate it.

And then the limit that produced the worst reporting mistake I have made.

Google omits low volume queries from the query report to protect user privacy. Those clicks count in your unfiltered totals but vanish the moment you filter. Everyone knows this. Almost nobody knows how big it gets.

On that top property, 94.6 percent of clicks came from queries Search Console will not name. I spent real time analysing its query mix and drawing conclusions about intent, and every one of those conclusions was drawn from about five percent of the data. The dashboard looked complete. It was not, and nothing in the interface told me so.

Check the anonymised share on any account before you trust a query level finding from it. Above 70 percent, your query analysis is a sample, and you should say so out loud to whoever you are reporting to.

Third party tools will not rescue you here either. Ahrefs, Semrush and the rest estimate organic traffic from rank positions and modelled click curves. They do not observe your visitors. Use them to generate a hypothesis. Confirm it in Search Console and Google Analytics, which count rather than estimate.

The related misread I see in almost every client conversation is treating rising impressions as visibility won. It usually is not. It is usually a wider, weaker query set, which pulls average position down and CTR with it, while nothing good has happened. That is exactly what my own property’s numbers look like, and I would rather show you that than a case study where everything went right.

What I would actually do on Monday

If you have read this far because your own chart is falling, here is the order I would work in, and it is deliberately unglamorous.

Spend the first twenty minutes proving there is nothing to fix. Check 12 September 2025, check the anomalies page, check whether anyone added a property. A meaningful share of the drops I get sent turn out to be reporting, and finding that out costs twenty minutes instead of a quarter.

Then spend an hour on the position column before you touch a single page. Not the clicks. Not the traffic graph in your analytics. The position column, split by query intent, compared year over year. That one hour decides which of the five prescriptions you follow, and following the wrong one is worse than doing nothing.

Only then open the content. And when you do, accept that some informational traffic is not coming back, because it is not a ranking you lost, it is a click the user no longer needs to make. The work moves to being the source the answer is built from, and to the commercial queries where clicks still exist. That is a different job from the one most teams are still resourced for, and pretending otherwise is how agencies bill for recovery work that was never going to recover anything.

If you want a second pair of eyes on a drop you cannot place, I do this for a living and I am reachable.

Frequently asked questions

What percentage of searches are zero click?

In the United States, 68.01 percent of Google searches ended without a click between January and April 2026, measured by SparkToro using Similarweb clickstream data, against 60.45 percent in 2024. Other studies report lower figures because they measure different things. Pew Research, tracking real browsing, found users clicked a traditional result on 15 percent of searches with no AI summary present and 8 percent where one was present. Treat any single figure as directional.

Why are zero click searches growing?

Three compounding reasons. Google has steadily added answer surfaces to the results page, from featured snippets in 2014 through knowledge panels and People Also Ask to AI Overviews in 2024 and AI Mode after it. AI Overview coverage expanded, from roughly 30 percent of tracked queries in February 2025 to roughly 48 percent a year later on BrightEdge’s measure, and it spread from informational into commercial queries, growing 71 percent on commercial intent SERPs between November 2025 and April 2026. And behaviour shifted with it: Pew found users ended their session entirely on 26 percent of visits where an AI summary appeared, against 16 percent where none did.

Do AI Overviews reduce clicks for every query?

No. The effect concentrates on informational queries and is much weaker on transactional and navigational ones. It is also not universally negative: Seer Interactive’s January 2025 data showed organic CTR slightly higher with an AI Overview present, 3.19 percent, than without it, 2.93 percent. The relationship inverted later in 2025, reaching 1.31 percent with against 3.97 percent without in December 2025, then partially rebounding to 2.36 percent by February 2026.

How do I know if AI Overviews are showing for my keywords?

Two ways, neither perfect. In Search Console, the generative AI performance report shows impressions from AI Overviews and AI Mode, live worldwide since 31 August 2026, but it gives impressions only and does not split the two. Manually, search your top queries in an incognito window and record which return an AI Overview. There is no AI Overviews filter in the search appearance dimension, so you cannot isolate them in the ordinary performance report.

Can I block my content from AI Overviews?

Yes, and this answer changed in 2026, so be careful with older articles. Search Console now carries a Search generative AI control under Settings that excludes your site from AI Overviews, AI Mode and generative AI features in Discover, without removing you from ordinary Google Search. Google rolled it out to all websites worldwide on 31 August 2026 and states it is not used as a ranking signal outside those generative AI features. Exclusion takes effect within one to two days.

Two things people get wrong. The older snippet controls, nosnippet, data-nosnippet and max-snippet, do limit what appears in AI features, but they are Search wide preview controls, so using them also kills your featured snippets and ordinary snippet text. And Google-Extended does not control AI Overviews at all: it governs whether your content trains and grounds Gemini models, and Google states explicitly that it does not affect inclusion or ranking in Google Search.

Worth knowing before you act: as of September 2026 Google’s own developer documentation on AI features has not been updated to reflect the new control and still describes robots.txt and snippet controls as the only instruments available. Two Google surfaces are out of step. The Search Console Help page is the current one.

Is a traffic drop with stable rankings always caused by AI?

No, and this is the most common false positive. Stable rankings with falling clicks can also come from a query mix shift that lowers average CTR without changing any ranking, from seasonality, from a logged Search Console reporting error, from comparing against preliminary recent data, or from a new property changing what is counted. Rule those out first. The combination you actually need is flat impressions, flat average position, falling clicks, and the loss concentrated on informational queries.

My impressions went up but my clicks went down. Is that AI?

Maybe, and this is the single most asked question on the forums, in almost exactly those words. Check three things in order. First, did your impression rise reverse on or around 12 September 2025, and was it desktop-weighted? Then it was rank tracker bots, not readers. Second, did average position get worse as impressions rose? Then you picked up a wider, weaker query set and your CTR fell arithmetically, with no behaviour change at all. Third, only if position held flat and the losses concentrate on informational queries, it is AI absorption. Worth remembering that people were asking this exact question on SEO forums in March 2022, two years before AI Overviews launched.