All of the Lab Journal
GEO / AI Visibility

What Zero-Click Search Means for Your Website Traffic

Baptiste Lacroix
Founder of MentionLab
BlueWritten with Blue
July 15, 2026

A zero-click search is a query that Google answers directly on the results page, through an AI Overview, a featured snippet, a knowledge panel, or a local pack, so the person searching never has to open a website. Bain & Company's February 2025 research puts the resulting drop in organic traffic at 15% to 25%, depending on the query type. That number is real, but it is not the whole story. The question worth asking in 2026 is not how to force the click back. It is which searches you can afford to give up on, and which ones you still need to win.

A zero-click search is a query that ends on the results page itself: the person searching gets an answer, a fact, or a completed action, and never clicks through to a website. Two different things happen inside that one definition.

In the first, an answer is enough: an AI Overview, a featured snippet, a knowledge panel, or a People Also Ask entry gives the searcher a complete response and the search stops there. In the second, the action happens inside Google itself: a call button, a set of directions, a shopping module. Neither sends a visitor to your site, but they close the loop for different reasons.

The results page used to be a waypoint. For a growing share of queries, it is now the destination. How an AI Overview is actually assembled, from source selection to summary generation, is a separate mechanism worth understanding on its own terms.

How much traffic does zero-click search actually cost you?

Zero-click search does not just change how people search. It changes how much of that traffic ever reaches your site, and the size of the drop is measurable.

Bain & Company's February 2025 research estimates that zero-click behavior cuts organic traffic by 15% to 25%. The same research finds that roughly 60% of searches now end without the person visiting any third-party website, and that about 80% of consumers rely on a zero-click answer in at least 40% of their searches. Together, those figures describe a search experience where getting an answer, not visiting a site, is now the default outcome for a large share of queries.

Pew Research Center's analysis of actual search sessions, published July 22, 2025, adds a second angle: when an AI summary appears above the results, the click-through rate on any link falls to 8%, compared with 15% when no summary appears. Pages with an AI summary present were also more likely to end the browsing session outright: 26%, compared with 16% for pages without one.

These are industry-wide estimates, not your numbers. Methodologies and panels differ, and the only measurement that tells you what is actually happening on your own pages is the one you run yourself. We cover how to measure what AI search actually sends to your own pages in a dedicated breakdown.

Why does Google say clicks are stable when your reports say otherwise?

Both statements are true at the same time, because they are not measuring the same thing.

Pew Research Center's study, conducted in March 2025 and published July 22, 2025, measured a real drop: the click-through rate per search falls from 15% to 8% once an AI summary appears. Google's own account, given by Liz Reid, VP and Head of Search, on August 6, 2025, describes something different: total organic click volume has stayed "relatively stable year-over-year," with click quality actually increasing.

Both can be accurate. A click-through rate that falls per search can still add up to a stable total if the number of searches keeps growing, and AI Overviews now sit in front of far more queries than a year ago. At the same time, the clicks that survive are the ones that happen after the summary has already answered the easy part of the question. The results page is now doing the filtering that used to happen in your opening paragraph.

That has a direct, practical consequence. A click-through rate that keeps falling while impressions keep climbing is not automatically a failure. It can be exactly what this shift looks like from inside Search Console. The metrics that survive a falling click rate are worth tracking on their own terms, separate from click-through rate alone.

If nobody clicks, what does a citation actually buy you?

A citation buys placement in front of an audience your blue link was never going to reach, not a session on your website.

The scale of that audience is not small. Alphabet's June 3, 2026 investor presentation put monthly AI Overviews users above 2.5 billion, up from more than 1.5 billion a year earlier in Q1 2025, and AI Mode above 1 billion monthly users. That is a placement, not a rounding error.

Being the source an AI Overview or a chatbot names buys three things: your brand in front of someone at the exact moment a decision is forming, a spot on the short list that survives the summary, and a seed for the branded search that follows once someone types your name into Google later. Where ChatGPT picks the sources it names, and what makes a page get selected as a cited source, are both worth understanding.

What a citation does not buy is just as important to say plainly. There is no session, no form fill, no last click you can point to in an attribution report. Pew Research Center found that only 1% of visits went to a source cited inside an AI summary itself, even when that source was named.

Treat the citation as a channel, not a consolation prize. Your paragraph is the message, the citation is the placement, and the branded search that follows is the response. A channel earns its keep on reach and on what it triggers next, not on whether the click happens in the same five seconds.

This article, Blue could have written it for you: content optimized for Google + AI, without you writing a single word.

Try mentionLAB

Which searches should you concede, and which should you defend?

Not every query deserves the same response. A short answer cannot replace a purchase decision, and a purely factual query never merited a full session, whether Google summarizes it or not.

Query typeWhat the SERP does with itWhat to publishWhat success looks like
ConcedeAnswered directly in an AI Overview or snippetA short, self-contained answer built to be lifted cleanlyThe cited source, not the click
DefendCompared or debated across several answersDepth and first-hand examples a summary cannot compressThe click still happens
CaptureReader knows the problem, not yet the fixA citation-worthy answer plus a reason to clickBoth the citation and the click

Concede

Concede queries are factual and definitional: a date, a formula, a conversion. A fact does not need a visit, it needs to be correct, so the right move is not to fight for the click. Write the answer so it gets lifted as the cited source: one sentence, verifiable, no throat-clearing before it. The payoff is brand impression, not traffic. Answer-first paragraphs a machine can lift cleanly are the mechanism that makes this work.

Defend

Defend queries carry a decision: a comparison, a trade-off, a price, an edge case that depends on specifics. A short summary cannot compress these without becoming wrong, which is why the click survives here. This is where depth, first-hand data, and a worked example earn their place, because a generic paragraph that could describe any competitor defends nothing. Interchangeable filler no engine has a reason to cite wins neither the click nor the citation.

Capture

Capture queries sit in the middle: the person knows their problem but not yet the solution. The goal is double: be the source the summary cites, and give the reader a reason to click that the summary cannot carry, original data, a worked number, a deeper answer than the short version could hold. This is the segment most sites leave on the table. FAQ sections built to be quoted, not skimmed, do a lot of this work at once.

This triage changes what gets published each month, not just what gets measured. Factual pages get shorter and tighter. Decision-stage pages get the first-hand depth a summary cannot replace. Content written to be cited in the first place, rather than retrofitted for AI after the fact, is where content strategy is heading over the next few years.

Frequently Asked Questions

Common examples include checking the weather, converting currency, finding a store's hours through a local pack, or reading a definition served by an AI Overview. A B2B example works the same way: a competitor's pricing tier often returns a complete answer on the results page itself, no click required.

How many Google searches are zero-click?

Bain & Company's February 2025 research estimates that around 60% of searches end without the person visiting any third-party website. Public estimates vary depending on the panel and methodology used, so treat any single figure as directional, and prioritize what your own Search Console data shows for your own queries.

Are zero-click searches increasing?

Yes, and the surface producing them keeps expanding. Alphabet's investor materials show AI Overviews growing from more than 1.5 billion monthly users in Q1 2025 to more than 2.5 billion by June 2026. That figure describes how many people use the feature, not a zero-click rate, but a surface that size is hard to ignore.

No. Google's Liz Reid stated on August 6, 2025 that the total volume of organic clicks sent to websites has remained relatively stable year-over-year, with click quality actually improving. What changed is which queries still earn a click, not whether clicks exist at all. Search did not end, it got more selective about which content wins one.

Paid ads buy a placement on the results page, not the status of a cited source inside an AI summary, and they face the same crowding from AI Overviews, local packs, and other features that organic listings do. Whether paid search fits your channel mix is a budget and strategy decision on its own, independent of the zero-click question.

Blue handles your SEO and your GEO. On autopilot.

You approve, she produces content optimized for Google + AI.

Join the Lab · 5-day trial

Read next

GEO / AI Visibility

Where ChatGPT Actually Gets Its Information

ChatGPT answers depend on two separate systems that work nothing alike: a training corpus frozen at a fixed cutoff, February 16, 2026 for the frontier models running today, and live web retrieval, which fetches current pages during the conversation itself, a capability every user has had since February 5, 2025. This article traces both paths using OpenAI's own 2026 documentation, including the specific crawlers behind each one, GPTBot for training and OAI-SearchBot for search, and the only training mix OpenAI has ever published in detail (GPT-3's, from 2020, not anything more recent). The second half turns practical: since a site cannot get into the training corpus on demand, it covers what actually makes a page retrievable and citable in ChatGPT's answers right now.

GEO / AI Visibility

Where SEO Is Headed Over the Next Few Years

SEO is not disappearing, but the exchange behind it is being repriced by numbers that can actually be checked. Google users click a normal search result 8% of the time when an AI summary appears, versus 15% when it does not, according to Pew Research Center (July 2025). Meanwhile, Google's crawlers pull far fewer pages per referred visitor than AI answer engines do: about 5.4 pages per visitor for Google versus more than 38,000 for Anthropic, per Cloudflare data from July 2025. This article separates what has been measured from what is being projected, names the mechanism behind the shift, and reports the contrary evidence too, including that Google's own Search query volume just hit an all-time high.

GEO / AI Visibility

sameAs Schema: How a Site Proves It Is Who It Claims

sameAs is the schema.org property that tells a machine which other pages, a Wikidata entry, a Wikipedia page, an official profile, describe the same entity as your site. Its purpose is entity disambiguation: without it, nothing separates your organization from every other one carrying a similar name. This article gives the selection rule missing from the pages currently ranking for the topic: a three-question test for deciding which URLs actually belong in the property, why reciprocity is a condition of validity rather than a nice-to-have, and a real implementation made on this site on June 14, 2026, including what was deliberately left out. It also covers, using Google's own documentation, why sameAs isn't an AI citation lever.