68% of Google Searches Now End Without a Click

The number that made me stop and re-read it: 68.01% of US Google searches ended without a click between January and April 2026. That is Similarweb clickstream data across desktop and mobile, published by SparkToro.

In 2024 the same measurement said 60.45%.

A 7.56 percentage point shift in two years does not sound dramatic until you flip it around. The share of searches producing at least one click fell 9.51 points, which is a 22.9% relative decline. Roughly a quarter of the clicks that Google used to hand out, gone, in twenty-four months.

And that 68% figure includes clicks to Google's own properties. Maps, YouTube, the shopping panel. The share going to the open web, the part that lands on your site, is smaller than the leftover 32% implies.

But the number is not really the point of this post. The number has been written about to death, usually by people selling you a solution to it. The thing I have not seen covered honestly is the second-order problem, which is worse:

You cannot see this happening in your own analytics. By design. And the tooling that claims to fix that costs more than most indie products earn.

Why Your Dashboard Looks Fine While You Lose

Here is the mechanic that makes zero-click loss so hard to notice.

When someone searches, reads an AI Overview, and does not click, nothing happens on your server. No pageview. No referrer. No entry in any log you own. The event that cost you a visitor produced exactly zero data on your side.

What you see instead is a slow, unremarkable decline in sessions with no single cause you can point at. Traffic is down 8% quarter over quarter. Nothing broke. Your rankings are the same. Search Console shows your impressions holding steady or even climbing while clicks slide.

That last pattern, impressions flat or up while clicks fall, is the clearest fingerprint of AI-driven loss you have access to for free, and most people never look at it because they look at the clicks graph and not the ratio. If your position has not moved and your impressions have not moved but your CTR has dropped by a third, an answer above you is eating the click.

The scale of that effect is not subtle. When AI Overviews appear, CTR on the results below drops by nearly 60%. And Google's AI Mode, the fully conversational surface, produces a zero-click rate around 93%, more than double regular AI Overviews. If AI Mode becomes the default rather than a tab, the numbers in this post age badly in the wrong direction.

The Second Blind Spot: Assistant Traffic That Never Touches Google

The zero-click statistic only measures Google. It does not measure the searches that never went to Google at all.

Somebody asks ChatGPT which tool they should use for X. The assistant names three products. Your product is either in that list or it is not. There is no impression, no ranking, no Search Console row. Nothing. If you are not named, the entire interaction is invisible to you, and it is invisible in a way that is qualitatively different from ranking eleventh, because at least ranking eleventh shows up in a report.

You get one weak signal: referral traffic. Visits with a referrer of chatgpt.com, perplexity.ai, claude.ai, copilot.microsoft.com. Those are worth segmenting out in your analytics right now if you have not, because that traffic converts unusually well. Somebody who arrives after an assistant recommended you by name is much further down the funnel than somebody who found you on a listicle.

But referral traffic only tells you about the times you won. It says nothing about the far larger number of times a competitor got named instead, which is the number you actually need.

I ran into this directly with roastme.gg. I put the site through ChatGPT's live search by hand to see whether it would show up for anyone not already looking for it by name. Doing that manually, one prompt at a time, is the crudest possible version of measurement, and it was still more informative than anything in my analytics. That is a bad sign for the analytics.

What Free Measurement Actually Gets You

Before spending money, here is the honest list of what you can do for nothing. It is more than most people do and less than you need.

Segment assistant referrals. Build a report filtering referrers for the major assistant domains. Track it weekly. This is your "am I being cited" proxy, biased toward whichever assistants pass a referrer at all.

Watch the CTR ratio in Search Console, not the click count. Pull position, impressions, and clicks per query. Sort by queries where position is stable and CTR has fallen. That list is your AI Overview exposure.

Check your server logs for assistant crawlers. GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and friends. Crawling is not citation, but never being crawled means citation is impossible. If they are not in your logs, check your robots.txt and your firewall rules before anything else, because plenty of people are blocking these by accident through a WAF preset.

Ask the assistants yourself, on a schedule. Write down fifteen prompts a real customer might type. Run them monthly in a fresh session with no history. Record which brands get named. This is tedious and unglamorous and it is genuinely the highest-signal free thing on this list.

That last one is the whole game, and the reason a tool category exists is that doing it by hand across four assistants and fifty prompts every week is not a thing a person will keep doing past week three.

The Tooling, and What It Costs

The AI visibility tracking category filled up fast. The shape of the offering is consistent: you give it a list of prompts, it runs them against the assistants on a schedule, and it tells you how often you get named and who gets named instead.

Rough pricing as of now:

• Otterly: from $25/month annually for 15 prompts tracked daily across AI Overviews, ChatGPT, Perplexity, and Copilot. $160/month for 100 prompts.
• Peec AI: from €89/month for 25 prompts, €199/month for 100.
• Profound: from $99/month for 50 prompts on ChatGPT, $399/month for 100 prompts across ChatGPT, Perplexity, and AI Overviews. Enterprise is custom.

Those prices are aimed at marketing teams, and for a marketing team they are trivial. For a solo founder doing $2,000 MRR, $199 a month to find out you have a problem is a hard sell, especially since none of these tools fix anything. They are thermometers.

That is the actual gap in this category, and it is worth naming plainly. Every tool here tells you that a competitor got named for prompt seventeen. None of them tell you which page to change or what to put on it. You are left holding a diagnosis and no prescription, at $200 a month, which is the same position a rank tracker left you in fifteen years ago except more expensive.

The Measurement Trap

There is a failure mode I want to flag because I have fallen into it.

Measurement feels like progress. Setting up dashboards, segmenting referrers, running prompt panels, building a spreadsheet of who gets named for what. All of that is real work and none of it changes a single answer any assistant gives.

The thing that changes an answer is changing a page. Assistants answer from a live search over crawled, indexed content. If your page does not say the thing, no amount of tracking makes it say the thing.

And the loop is slow. A page you rewrite today has to be crawled and indexed before it can be cited at all, and realistically you are looking at three to four weeks before an assistant's answer could plausibly move. Anyone promising you a shift in two weeks is either lucky or lying.

So the measurement is only worth what you do with it, and the honest sequence is: measure enough to know which prompts you are losing, then spend the rest of your time on the pages, then measure again to see if it worked. One cycle is a month. That is the natural cadence of this work and it is much slower than the daily dashboards imply.

What Actually Moves an Assistant's Answer

This is where it connects to everything else. The pages that get cited are the pages that contain something an answer engine must point at rather than paraphrase, which is the same rule that now governs programmatic SEO and the same rule I laid out in the generative engine optimization guide.

Concretely, the things that keep showing up on cited pages:

Direct answers, stated plainly, near the top. An assistant extracting a claim needs a sentence it can extract. Three paragraphs of throat-clearing before the answer means the answer does not get taken.

Specific numbers with sources. "Faster" gets paraphrased away. "22% more requests under load" gets quoted.

Explicit comparisons. If someone asks which tool to use, the pages that get cited are the ones that actually name alternatives and say when to pick them. Most product pages refuse to mention competitors, which is exactly why competitors' pages get cited instead.

Structure that survives extraction. Headings that match real questions, FAQ blocks, tables. Not because schema is magic but because a table is unambiguous and a wall of prose is not.

Being named somewhere else. Assistants pull from live search, so pages about you that you do not control matter enormously. This is the part where directory listings and third-party mentions do real work, not for the link juice, but because they are corroborating sources an assistant can find.

None of this is exotic. It is mostly "write the answer down clearly and be mentioned elsewhere," which was good advice in 2015 too. What changed is that the penalty for not doing it went from ranking lower to not existing.

The Part I Keep Coming Back To

The 68% number gets reported as a search story. It is not. It is a distribution story.

For a decade, the deal was that if you wrote something good enough and structured it well enough, Google would send you strangers. That deal was never guaranteed and it is now visibly ending, and the replacement is not another channel you can optimize into. Being named inside an answer you cannot see is not a channel. It is a mention.

Which loops back to something I have written more than once: building is easy and distribution is the moat. What is happening to search is a specific case of the general rule. The distribution channels that were legible and gameable are becoming illegible, and the things that still work are the slow ones. Having a real audience. Being the source somebody else quotes. Owning a dataset. An email list that does not depend on anyone's ranking algorithm.

Measure the AI answers, absolutely. Just do not mistake the dashboard for the work.

Frequently Asked

What percentage of Google searches end without a click? 68.01% of US Google searches ended without a click between January and April 2026, per Similarweb clickstream data published by SparkToro, up from 60.45% in 2024.

Can I see AI Overview traffic loss in Google Analytics? No. A search that ends in an AI answer never reaches your server, so it produces no data on your side. The closest free signal is Search Console: stable position and impressions with falling CTR.

How do I track whether ChatGPT mentions my brand? Manually, by running a fixed set of customer-style prompts in fresh sessions on a schedule and recording who gets named. Automated tools do the same thing weekly, from roughly $25 to $399 a month depending on prompt volume and engine coverage.

Does traffic from ChatGPT show up as a referral? Often, yes. Segment referrers like chatgpt.com, perplexity.ai, and claude.ai in your analytics. That traffic tends to convert well, but it only shows the times you were cited, never the times you were not.

How long does it take for a page change to affect AI answers? Realistically three to four weeks. Assistants answer from live search, so a rewritten page has to be crawled and indexed before it can be cited at all.

What is Google AI Mode's zero-click rate? Around 93%, more than double the rate for standard AI Overviews.

Pick fifteen prompts your customers would actually type, run them tonight, and write down who gets named. Whatever you find will be more useful than another quarter of watching sessions drift down for no visible reason.