TL;DR: The click loss from AI Overviews is real and it has been measured, but it is not evenly distributed. It lands hardest on informational queries and barely touches the commercial and local searches most small businesses actually sell on. The fix is to stop treating clicks as the only unit of visibility and start measuring citations, which are already trackable in tools you have.
The one-sentence answer: Organic click-through rate falls by roughly a third to a half on searches where an AI Overview appears, that appearance is concentrated in informational queries, and the correct response is to move your content down-funnel and add citation tracking rather than to publish more of what AI now answers for free.
Every few weeks someone sends me a screenshot of their analytics with a red line going down and asks whether SEO is dead.
The line is usually real. The conclusion usually is not.
There is now enough published measurement to say what is actually happening with reasonable precision, and to say which parts of it should worry a service business and which parts should not. This post is the numbers, where they came from, and what changes as a result. The theory of getting into AI answers is covered in our generative engine optimization guide. This is about what happens to your traffic while you do it.
What does the click data actually show?
Two independent studies, using different methods, point the same direction. Pew Research Center instrumented the browsers of 900 U.S. adults and recorded 68,879 Google searches during March 2025, of which 12,593 produced an AI summary. On pages with an AI summary, users clicked a traditional search result in 8% of visits. On pages without one, they clicked in 15% of visits. Users clicked a link inside the summary itself in 1% of visits (Pew Research Center, July 22, 2025).
That is close to half the clicks, measured on real browsing rather than on rank-tracking estimates.
Seer Interactive ran the other kind of study, aggregating 5.47 million queries and 2.43 billion organic impressions across 53 brands from January 2025 through February 2026. Their February 2026 figures: organic CTR of 2.36% on searches with an AI Overview present, against 3.82% without. Paid CTR moved too, 16.21% against 21.85% (Seer Interactive, April 24, 2026).
Two studies, two methods, same shape. The relative drop is somewhere between a third and a half.
Which queries are actually losing the clicks?
This is the part that gets left out of the panic coverage, and it is the part that determines whether any of this is your problem.
AI Overviews do not appear evenly. In Seer's February 2026 data, 36% of informational queries returned an AI Overview. On commercial queries it was 8%, and on transactional queries 5%.
Read that again with your own search terms in mind. "How does a silicone roof coating work" is an informational query and it is exposed. "Commercial roofing contractor near me" is not the same animal at all. The mechanics of which queries trigger the feature are in our guide to how Google AI Overviews work. The searches that produce quotes and phone calls are, so far, the searches where Google is least likely to answer on your behalf.
So the honest framing for a service business is narrower than the headlines: you are probably losing clicks at the top of your funnel and keeping them at the bottom. If your blog exists to catch people three months before they buy, that is a real loss. If your money pages carry the local and commercial terms, those are mostly intact.
Is Google right that traffic is stable?
Google's position deserves to be stated properly rather than dismissed, because it is the strongest counterargument to everything above.
Liz Reid, VP and Head of Google Search, published it on August 6, 2025: "Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year." She adds that average click quality has risen, defining a quality click as one where the user does not quickly click back, and says the third-party reports showing dramatic declines are "often based on flawed methodologies, isolated examples, or traffic changes that occurred prior to the roll out of AI features in Search."
Google sees the whole system. Nobody else does. That is a genuine advantage and the criticism of sampling-based studies is fair on its face.
Here is why it still does not settle the argument: the post contains no numbers. Not one figure, percentage, or time period that anyone outside Google could check. "Relatively stable" and "slightly more quality clicks" are directional claims, and a directional claim with no data behind it cannot be tested, only believed or not. Pew published its sample, its date range, and its counts. Seer published its query volume and impression counts. Google published adjectives.
Both things can also be true at once. Aggregate click volume across the entire web can hold steady while the distribution underneath it shifts away from pages that answer simple questions. Aggregates hide exactly that kind of movement, which is why your own analytics can fall while Google's total stays flat.
What happens if you do get cited?
Getting cited helps, and it does not restore what was lost.
Seer's February 2026 breakdown on informational queries: being cited in the AI Overview delivered 120% more organic clicks per impression than going uncited on the same result page. That same cited position still produced 38% fewer clicks than the query would have delivered with no AI Overview present at all.
Sit with the second half of that. Winning the citation is worth roughly double, and still lands you meaningfully below where you were before the feature existed. Any advice that presents AI citation as a straight replacement for the traffic is selling something. It is a partial recovery and a brand impression, which is worth pursuing on its own terms.
What should you stop measuring?
Stop reading blended organic sessions as a health metric. It now mixes two populations that are moving in opposite directions, and the blend tells you nothing actionable.
Split them instead. Group your informational URLs and your commercial URLs and look at them separately, which is the first move in our AI SEO audit checklist. Most of the "traffic collapse" stories resolve, on inspection, into a normal commercial page set and an informational set that fell off a cliff. That is a content-mix problem with a known fix, and it reads as an existential crisis only while the two are averaged together.
Also stop treating average position as progress. Across our own site's trailing 90-day query export to August 9, 2026, 44 queries sat at position 20 or better. They accounted for 333 impressions, 1.3% of all query impressions, and produced zero clicks. Impressions were growing the whole time. Position and impressions can both improve while clicks stay at zero, because ranking on page four of a query nobody clicks is not a business outcome.
What should you measure instead?
Three things, all available without new software.
Citations. Bing Webmaster Tools has reported them since its AI Performance report entered public preview on February 10, 2026. It shows total citations, average cited pages, and the grounding queries the AI used when it retrieved you. Google's Generative AI performance report in Search Console covers AI Overviews and AI Mode, but reports impressions only, with no clicks and no queries. The tracking detail is in our guide to tracking AI citations.
Assisted conversions from AI referrers. In GA4, traffic from ChatGPT, Perplexity and Copilot lands in Referral by default. A custom channel group fixes it, and because custom channel groups apply retroactively, you get your history back the day you build it.
Branded search volume. If citations are doing their job, more people arrive already knowing your name. Branded query impressions in Search Console is the closest available proxy for whether being quoted is compounding.
We publish our own version of this every month, including the months where the numbers are bad. In our fourth month, Bing AI citations went from a 66.0 per day average to 14.4, a 78% drop, and held there for eleven days before recovering to 60.7. The single-day move on July 27, 2026 was 50 citations across 12 pages down to 2 across 1. The obvious explanation was a bulk metadata change we had shipped. It was wrong: the same collapse and recovery hit three unrelated domains on the same dates, and one of them had not been touched at all. A single-property view produces confident, wrong causal stories. Check a second property before you blame your own deploy.
What does this change about what you publish?
Less than the panic suggests, and in one specific direction.
Google's own guidance, updated July 10, 2026, is blunt about the things vendors are currently selling. Structured data is not required for generative AI features and there is no special schema to add. Machine-readable AI text files, llms.txt included, are not used by Google Search. There is no requirement to break content into small pieces. The stated eligibility bar is that the page is indexed and eligible to appear with a snippet.
What changes is the brief, not the technique. A post that exists to define a term is now competing with a free answer at the top of the page, and it will lose. A post that carries a judgment, a number nobody else has, or a decision the reader has to make cannot be pre-answered, because the AI has nothing to summarize until you publish it. That is the same content advice as always, with the low end of the market removed.
Practically: fewer definition posts, more comparison and decision posts, and put a real number in each one.
Where does this advice fail?
In three places, and they are worth naming.
If you are a publisher monetizing informational traffic with ads, the reframe above does not help you. Your product is the click. Moving down-funnel is not an adjustment, it is a different business, and nothing in this post solves it.
If your category is genuinely informational at the point of sale, medical questions and insurance among them, the exposure follows the query, not the industry. Check your own top queries against the intent split rather than assuming a service business is safe.
And these figures have a shelf life. The Pew data is March 2025 browsing. The Seer figures are February 2026. AI Overview trigger rates have moved every quarter since launch, and any number in this post could be wrong by the time the feature set changes again. Re-run the check rather than trusting a percentage you read once.
Frequently Asked Questions
How do I tell whether an AI Overview is appearing on my terms?
Search your top twenty commercial queries in an incognito window from the location you serve, and record which ones return an AI Overview. It takes about twenty minutes and it is more reliable than any category-level average, because trigger rates vary by query wording, not just by industry.
Should I block AI crawlers to protect my content?
Check which crawler you are actually blocking first, because the vendors run separate agents for training and for search retrieval. GPTBot and ClaudeBot are training crawlers, and blocking them costs no citations. OAI-SearchBot, PerplexityBot, and Claude-SearchBot are the retrieval agents, and blocking those removes you from the answers. In our site audits the blanket block is almost never a decision anyone made: it is a security plugin or CDN toggle that shipped switched on, and it takes the search agents down with the training agents.
Does paid search fix the gap?
Not cleanly. Paid CTR fell in the same Seer dataset, from 21.85% to 16.21% on queries with an AI Overview present. Budget can buy back volume, at a higher cost per click than the same query commanded before.
How long before this settles?
There is no reason to expect it to. The trigger rate, the interface, and the citation behavior have all changed repeatedly since launch. Build the measurement so that you can see the next shift when it happens, rather than waiting for a stable state to plan around.
The uncomfortable part of this is that the reporting has to get more granular before it gets more reassuring. Blended organic traffic will keep looking worse than your business is doing, and the split view is the only thing that tells you which half needs work.
If you want that split built properly, with citation tracking alongside it, see ClickWerxs AI SEO or get in touch.
Sources
- Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results" — 8% vs 15% click rate; 1% clicked a link within the summary; 26% vs 16% session ends; 900 U.S. adults, 68,879 searches, 12,593 with an AI summary, browsing data March 1-31, 2025. pewresearch.org (July 22, 2025; retrieved 2026-08-22)
- Seer Interactive, "AIO Impact on Google CTR: 2026 Update" — February 2026 organic CTR 2.36% with AIO vs 3.82% without; paid 16.21% vs 21.85%; AIO appears on 36% of informational, 8% of commercial and 5% of transactional queries; cited vs uncited on the same result page +120% organic clicks per impression, still -38% vs no AIO; 53 brands, 5.47M queries, 2.43B organic impressions, January 2025-February 2026. seerinteractive.com (April 24, 2026; retrieved 2026-08-22)
- Liz Reid, VP and Head of Google Search, "AI in Search is driving more queries and higher quality clicks" — total organic click volume "relatively stable year-over-year"; quality-click definition; criticism of third-party methodologies. No figures published. blog.google (August 6, 2025; retrieved 2026-08-22)
- Google Search Central, "Google's Guide to Optimizing for Generative AI Features on Google Search" — structured data not required; llms.txt and AI text files not used by Google Search; no content-chunking requirement; eligibility requires indexing and snippet eligibility. developers.google.com (last updated July 10, 2026; retrieved 2026-08-22)
- Bing Webmaster Blog, "Introducing AI Performance in Bing Webmaster Tools Public Preview" — total citations, average cited pages, grounding queries. blogs.bing.com (February 10, 2026; retrieved 2026-08-22)
- Google Search Console Help, "Generative AI performance report (Search)" — impressions only, AI Overviews and AI Mode combined; requires sufficient impressions to populate and excludes sites opted out of generative features. support.google.com (retrieved 2026-08-22)
- OpenAI, "Bots" —
OAI-SearchBotsurfaces websites in ChatGPT's search features;GPTBotcrawls content for model training. developers.openai.com (retrieved 2026-08-22) - Perplexity, "PerplexityBot" — surfaces and links websites in Perplexity search results. docs.perplexity.ai (retrieved 2026-08-22)
- Anthropic, "Does Anthropic crawl data from the web?" —
Claude-SearchBotfor search result quality;ClaudeBotfor training. support.claude.com (retrieved 2026-08-22) - ClickWerxs first-party analytics — 44 queries at position 20 or better across the trailing 90-day query export to August 9, 2026, 333 impressions, 1.3% of query impressions, zero clicks; Bing AI citations 66.0/day (July 18-26) falling to 14.4/day (July 27-August 6) and recovering to 60.7/day (August 7-9), replicated across three unrelated domains. Published in full in our month four results. Operator data, on record.
ClickWerxs sells AI SEO services built on the methodology described in this post. First-party citation and traffic figures are our own past results, not a promise of outcomes for any engagement. Third-party figures reflect the publication dates cited and change frequently.
Kaleb Dickhaut — Founder, ClickWerxs. Kaleb built ClickWerxs from the ground up, from payment processing ISO to the Command Center platform to the AI SEO methodology the blog runs on. He has onboarded hundreds of small businesses onto payment and CRM systems. linkedin.com/in/kaleb-dickhaut