The End of the Search Click and What Comes Next

The search click is not disappearing all at once, but the majority of Google searches now end without one. In the first four months of 2026, SparkToro's analysis of Similarweb clickstream data found that 68.01% of US Google searches ended without any click to the open web, up from 60.45% in 2024. Fewer than one in three searches now sends a visit to an external site. For publishers who built their traffic on ranking well and being clicked, this is the central shift of the decade. This article sets out what the evidence actually shows, why AI Overviews are driving most of the near-term damage, what you should measure instead of raw clicks, whether an AI citation without a visit is worth anything, and how to prepare your site for a web where being quoted matters as much as being visited.

Is the search click disappearing?

Not entirely, and not evenly, but the direction is clear and the pace is accelerating. The click is now the minority outcome on Google, and its value is falling at every ranking position. Three independent lines of evidence point the same way.

First, the top-line trend. SparkToro's clickstream work puts the share of US Google searches ending without a click to the open web at 68.01% in early 2026, up from 60.45% two years earlier. That is a structural move, not a blip.

Second, the mechanism. Google AI Overviews, the AI-generated summaries at the top of many results pages, measurably suppress clicks. The Pew Research Center tracked 68,879 unique Google searches from 900 US adults during March 2025, of which 12,593 produced an AI summary. On pages where an AI summary appeared, users clicked a traditional result link on 8% of visits, against 15% on pages without one, roughly half. Only 1% of visits to AI-summary pages produced a click on a link inside the summary itself. The AI answer does not merely redistribute clicks, it removes them.

Third, the trajectory. Ahrefs' study of 300,000 keywords, built on aggregated Search Console data, found that position-1 click-through on AI Overview keywords fell 58% between December 2023 and December 2025, dropping from 0.073 to 0.016. That is up from the 34.5% reduction the same team measured in April 2025. The drag runs down the whole results page, with position 2 off 50.8% and position 10 off 19.4%. The click is not vanishing, but the value of a ranking is being steadily eroded, and the erosion is getting worse over time.

There is corroborating causal evidence too. A pre-registered randomised field experiment of 1,065 US Chrome users, run over January and February 2026 and reported by Search Engine Journal, found that AI Overviews cut organic clicks by 38% on triggered queries and pushed the zero-click rate from 54% to 72% when an AI Overview was present. Because that is a controlled experiment rather than an observation, it strengthens the case that AI summaries are causing the decline rather than merely coinciding with it.

What is a zero-click search?

A zero-click search is any search that ends without the user clicking through to a website in the open web. The query is answered, or abandoned, on the results page itself. This is not new. Weather boxes, calculators, sports scores and knowledge panels have absorbed clicks for years. What is new is the scale, and the fact that AI summaries now answer far more complex, article-shaped questions that used to require a visit.

The Pew data captures this cleanly. Users ended their browsing session entirely on 26% of pages that showed an AI summary, against 16% of pages without one. In other words, the AI answer does not just prevent a click to your site, it often satisfies the user so completely that they stop searching altogether. The session ends, and there is no second-choice result waiting to catch the visit.

It helps to separate two AI surfaces on Google. AI Overviews are the summaries injected into the standard results page. AI Mode is Google's fuller conversational search experience, which is even more zero-click than Overviews, with the great majority of sessions ending without an external click. But AI Mode is still small: SparkToro found it was entered on only 0.34% of searches in early 2026. AI Overviews, by contrast, appear on more than 20% of searches. So while AI Mode is the more extreme long-term threat, the near-term revenue damage comes overwhelmingly from AI Overviews, and that is where publishers should focus attention today.

If clicks fall, what should publishers measure instead?

The uncomfortable answer is that the metric most publishers rely on, Search Console clicks, is becoming both incomplete and misleading. It is incomplete because it never sees AI referral traffic. It is misleading because AI Overview and AI Mode performance is folded into ordinary search results with no way to isolate it. The fix is to measure three separate things: crawler activity, AI referral value, and citations. Each lives in a different place.

Start with what Search Console cannot show you. Google Search Console does not break out AI Overview or AI Mode performance. Clicks and impressions from both are counted inside the standard "Web" search type with no dedicated filter, so you cannot cleanly separate AI-influenced search performance from ordinary blue-link performance. A falling click count in GSC could be AI Overviews eating your clicks, or it could be a ranking drop, and the tool will not tell you which.

Now consider what GSC never sees at all. When someone clicks a citation inside ChatGPT, Perplexity or Claude, that visit arrives as a referral from the AI product's own domain, such as chatgpt.com or perplexity.ai. It did not come through Google Search, so it never appears in Search Console. This is a genuinely new and separate channel, and if you only watch GSC, it is invisible to you.

GA4 recognises it, but only partially. Google added a native "AI Assistant" channel to GA4's default channel group in 2026, keyed on the medium "ai-assistant". That is progress, but per Digital Applied's analysis it recognises only ChatGPT, Gemini and Claude out of the box. Perplexity, Copilot and others stay buried in Referral, or in Direct when the referrer is stripped, unless you build a custom channel-group regex to catch them. Left alone, GA4 undercounts AI traffic.

Given all that, here is a more honest measurement stack. There are three distinct AI events, and each is visible in a different place.

The practical shift is to stop treating publish count or raw clicks as the scoreboard. Measure net clicks, AI referral volume and its on-site value, and crawler activity together. A page can be quietly doing a great deal of work inside AI answers while its Search Console clicks fall, and you will only see the full picture if you look in all three places.

If you want a clean, Plausible-style read of your existing GA4 with an AI tab that surfaces AI referrals by engine, which crawlers are taking your content, and what AI-referred visitors earn against your site average, you can try Ramprt free on your own GA4. It reads your existing data read-only, with no new tracking script, so you can see the AI picture without instrumenting anything.

Does being cited by AI without a click have value?

Sometimes, but far less than a click, and you should not assume otherwise. The Pew figure is the sober anchor: only 1% of visits to AI-summary pages produced a click on a link inside the summary. A citation that does not generate a visit is the common case, not the exception. If your business model depends on ad impressions or on-site conversions, a citation with no click delivers no revenue directly.

That does not make citations worthless. A citation is a brand impression. Being named as the source in an AI answer builds authority and recognition, and it is the prerequisite for the small share of clicks that do come through. When AI referral clicks do arrive, they are worth measuring on their own terms, because AI-referred visitors often behave differently from search visitors, and their value against your site average is one of the more useful numbers you can track. Value citations for what they are, brand and pipeline for the occasional click, rather than pretending they replace lost search traffic pound for pound.

The strategic implication is uncomfortable but clear. If AI answers increasingly satisfy the query without a visit, a purely ad-funded model is structurally exposed on exactly the queries most likely to trigger an AI summary. The queries where you still earn a click are the ones worth understanding, protecting and expanding.

How do I prepare my site for a citation-first web?

Preparation splits into two jobs: deciding which AI systems may use your content and for what, and then measuring the traffic and citations you get. The first job is more fiddly than most publishers expect, because AI crawlers do not behave like a single Googlebot.

Understand the three crawler roles per vendor

Each major AI vendor runs distinct crawlers for distinct jobs: a training crawler, a search-index crawler, and a live user-fetch agent. Blocking the wrong one can remove you from AI search answers while leaving you exposed to training, or the reverse. The pattern is consistent across vendors.

The key detail from OpenAI's own documentation is that OAI-SearchBot surfaces your site in ChatGPT search, so blocking it removes you from ChatGPT's search answers. If your goal is to be cited, that is usually the opposite of what you want. The full user-agent for the training crawler is GPTBot/1.3. Perplexity follows the same split, with PerplexityBot handling indexing and Perplexity-User fetching pages on demand during a live query. Anthropic runs ClaudeBot for training, Claude-SearchBot for search quality and Claude-User for user-initiated fetches, and states that all of them honour robots.txt.

Know where robots.txt stops working

Robots.txt is the right tool for training and indexing crawlers, but it has a real coverage gap. OpenAI explicitly states that because ChatGPT-User visits are triggered by a human request, robots.txt rules may not apply to it. The same logic covers Perplexity-User and Claude-User. So blocking the training and index crawlers via robots.txt does not reliably stop on-demand fetches when a user asks the assistant about your page. If you genuinely need to stop those live fetches, you need IP or WAF rules, not just a robots.txt line.

Do not rely on Google-Extended to get out of AI Overviews

This is one of the most common and costly misunderstandings. Google-Extended is not a crawler and not a user-agent. It is a robots.txt product token, introduced in September 2023, that only governs whether your content is used to train Gemini apps and Vertex AI, including grounding. Per Search Engine Land, it manages training use only. Google has since stated that Google-Extended is not a Search ranking signal, and that AI Overviews are built from the live Google index crawled by Googlebot. So disallowing Google-Extended does not remove you from AI Overviews. The only way out of Google's index, and therefore its AI answers, is to block Googlebot, which also removes you from ordinary Search. For most publishers that is a non-starter.

Verify crawlers properly, because user-agents are spoofable

A user-agent string on its own proves nothing, because anyone can send it. There are two verification paradigms. The stronger is IP allow-listing against published machine-readable JSON. OpenAI publishes IP-range files at openai.com/gptbot.json, openai.com/searchbot.json and openai.com/chatgpt-user.json, and Perplexity publishes equivalents for its bots. Anthropic publishes a bot list too, but recommends robots.txt over IP allow-listing and warns that IP blocking may not persistently guarantee an opt-out. Where no authoritative IP list exists, reverse DNS or forward-confirmed DNS is the fallback. Whichever method you use, verify the user-agent and the source IP together. Checking one without the other tells you very little.

A practical sequence

Putting it together, here is a defensible order of operations for a publisher who wants to be cited, keep earning, and stay in control.

1. Decide your policy per role, not per vendor. Most ad-funded publishers want to stay in the search-index and live-fetch layers, since those produce citations and referral clicks, while making a separate, deliberate choice about training crawlers.
2. Set robots.txt for the training and index crawlers you want to exclude, using the exact tokens above.
3. If you need to stop live user fetches, add verified IP or WAF rules, because robots.txt will not reliably hold them.
4. Verify every AI request by user-agent plus IP against the published JSON lists before you trust or block it.
5. Instrument measurement across all three layers: server logs for crawler hits, GA4 with a custom channel-group regex for AI referrals beyond ChatGPT, Gemini and Claude, and prompt monitoring for citations that never convert to a visit.

The through-line is the same as the measurement section. The click is no longer the whole story, so your controls and your analytics both need to account for crawling, citation and referral as separate things.

What comes next

The realistic outlook is a web where being the source an AI trusts becomes as important as ranking first, and where a shrinking, higher-intent stream of clicks sits alongside a larger, mostly invisible stream of citations. The click is not dead. But it is now the minority outcome on Google, it is worth less at every position than it was two years ago, and the tools most publishers use to track it were built for a world that is ending. The publishers who adapt will be the ones who measure citations and AI referrals as first-class metrics, who make deliberate per-crawler decisions instead of blunt blocks, and who stop mistaking a healthy Search Console dashboard for a healthy business.

For a deeper look at the referral side of this shift, see our guide to measuring AI referral traffic in GA4, and for the crawler controls, our walkthrough of setting up robots.txt for AI crawlers.