What is a zero-click search?

A zero-click search is a search that ends on the results page, with the user getting what they needed without clicking through to any website. It happens through AI Overviews, featured snippets, knowledge panels, local map packs, calculators, weather cards and the growing set of answers Google assembles in place. The important nuance for measurement is that a zero-click search is not a failed search. The user was served, and in many cases they were served using your content. What changed is not whether your page did work, but whether that work leaves a footprint in your analytics. A click is a measurement event as much as it is a business event, and when the click disappears the business effect can continue while the measurement event does not. That gap is the entire problem, and treating a zero-click search as lost demand rather than as unmeasured demand is what leads teams to cut the exact content that is still doing the work.

How do you measure the impact of zero-click search?

You measure it indirectly, using several weak signals that together are reliable even though none of them is conclusive alone. We use five. First, the divergence signal in Search Console: query-level impressions holding flat or rising while click-through rate falls, which tells you an answer is being served above you. Second, branded search volume, which is the closest available proxy for demand that was absorbed on the results page and returned later as someone searching your name. Third, direct and unattributed traffic landing on deep commercial pages rather than the homepage, which is what a remembered brand looks like in GA4. Fourth, self-reported attribution on the enquiry form, which is the only signal that captures a prospect who read about you inside an AI answer and never visited at all. Fifth, a manual AI citation panel that records whether you are actually being quoted. No single one of these proves anything. Read together over a full quarter, they separate a brand being cited without clicks from a brand quietly losing relevance.

Does a drop in organic traffic always mean zero-click search is to blame?

No, and assuming it does is the most expensive mistake available here, because it converts a fixable problem into a comfortable story. Zero-click absorption has a specific signature: impressions flat or up, click-through rate down, and the decline concentrated in informational and definitional queries rather than commercial ones. If your impressions are also falling, that is not zero-click, that is a ranking or indexing problem and it needs ordinary diagnosis. If your decline is concentrated on commercial or bottom-of-funnel queries, that is competitive or intent loss, and it should worry you a great deal. If your enquiry volume fell in step with sessions, you have a real demand problem and no measurement framework will make it pleasant. The framework is only useful once the signature actually matches. Check the signature before reaching for the explanation.

Why do impressions stay flat while clicks fall?

Because impressions count the presence of your result on the page and clicks count the journey off it, and only the second one is affected when an answer is placed above your listing. When an AI Overview or featured snippet resolves the query in place, your result is still rendered, still counted as an impression, and still positioned where it was, but the user's need is satisfied before they reach it. This produces the pattern most teams first notice as an unexplained click-through rate decline on pages whose rankings never moved. It is worth being precise about what this does and does not tell you. Flat impressions with falling clicks confirms that something is intercepting the click. It does not, by itself, confirm that you were the source that the Overview used, and the two are commonly conflated. Confirming citation requires actually looking at the results page for those queries, which is why a manual panel remains necessary.

What percentage of searches are zero-click?

Published estimates vary widely and you should treat any single precise figure with suspicion, because the number depends heavily on what the study counted. Bain's 2025 research found that roughly 80 percent of consumers rely on zero-click results in at least 40 percent of their searches, which is a statement about user behaviour rather than a share-of-queries figure. Other studies count the proportion of all searches ending without a click and land anywhere from the high fifties to the high sixties depending on whether they include navigational searches, mobile versus desktop, and whether clicks to Google's own properties count as clicks. The practical answer is that the aggregate percentage does not matter much for your decisions. What matters is the zero-click rate for your specific query set, which you can approximate from your own Search Console data by tracking click-through rate at stable ranking positions over time. Your categories may be far above or far below any published average.

Should we stop optimising for informational keywords?

No, but you should stop valuing them the same way, and you should be honest about which ones are still earning their keep. The correct response is not abandonment, it is reclassification. Informational content in a zero-click environment does two jobs it did not do before: it is the raw material that AI answers are assembled from, which is how your brand name reaches a buyer who never visits, and it is the depth signal that makes you a plausible source on the commercial queries that still convert. What changes is the yardstick. Judging a definitional post by the sessions it generates will mark almost all of them as failures, and cutting them removes the substrate your citations are built from. Judge them instead by whether they get cited, whether the brand appears in the answers your buyers are actually asking, and whether branded search is rising. Some informational content genuinely is dead weight, and a content audit should still remove it. The distinction is whether the piece is being used or ignored, not whether it is being clicked.

How do you report zero-click value to a client or a CFO?

Not by inventing a monetary figure for impressions, which is the temptation and is indefensible the first time anyone interrogates it. What holds up is a reporting structure that separates three things and never blends them. First, the demand layer: impressions, query coverage, citation presence and branded search volume, reported as visibility and explicitly labelled as leading indicators rather than outcomes. Second, the traffic layer: sessions and click-through rate, reported honestly including the decline, with the query mix behind the decline shown so the audience can see whether it came from definitional or commercial terms. Third, the outcome layer: enquiries, qualified pipeline and revenue, which is the only layer anyone should be judged against. The argument you are making is that layer one and layer three moved together while layer two moved against them, and that is a defensible position precisely because you showed the decline rather than hiding it. A CFO is rarely hostile to a leading indicator. They are hostile to a leading indicator presented as a result.

Which content formats get pulled into zero-click answers most often?

Formats that are structurally easy to lift and that carry something specific enough to be worth lifting. In what we see, the recurring pattern is a self-contained block of roughly forty to sixty words that answers one question directly, sitting immediately beneath a heading phrased close to how a person would ask it, and containing at least one checkable particular: a number, a range, a named constraint, a date, a documented step, or an explicit exception to a general rule. Comparison tables with stated criteria and ordered process lists both perform well for the same reason, which is that they survive extraction without their surrounding paragraphs. What consistently underperforms is a well-argued narrative that requires three paragraphs of context before the answer appears, and generic best-practice advice that four other sources state identically. The first has no liftable unit. The second gives the synthesiser no reason to prefer you.

Do zero-click impressions actually turn into revenue, or is that just a convenient story?

It is a convenient story that also happens to be partly true, and the honest position is to hold both halves. The mechanism is real: a buyer who reads your name inside an AI answer, forms an impression, and searches your brand two weeks later has been influenced by content they never clicked. We see this in self-reported attribution, where a meaningful share of enquiries name a source the visitor never landed on, and in branded search rising while non-branded clicks fall. What we cannot do is close the loop deterministically. There is no join key between an impression and an enquiry, so anyone offering you a clean attributed number for zero-click revenue is modelling, not measuring. The defensible claim is directional: across the accounts we run, the correlation between citation presence and enquiry volume has been consistent enough that we act on it, while remaining unwilling to put a rupee figure on a single impression. Treat it as a channel you can see the shape of but not weigh precisely.

How long should you wait before concluding zero-click search is helping rather than hurting?

One full quarter at minimum, and two before you make an irreversible decision such as cutting a content line. The reason is that the two halves of the pattern move on different clocks. The traffic decline appears almost immediately, because an answer placed above your result changes click-through the day it appears. The compensating effects are lagged: branded search rises only after enough people have encountered and remembered the brand, self-reported attribution only accumulates as enquiries arrive, and pipeline shows up on whatever your sales cycle happens to be. Judging the pattern at four weeks therefore guarantees you see the loss and none of the gain, which is exactly when most teams panic and cut. The corollary matters too. If you have run two clean quarters and branded search is flat, citations are absent and enquiries fell in line with sessions, then the reassuring story has failed its own test and you should treat the decline as real.

Zero-Click Search Cost Us Traffic but Won Us Clients

The meeting that produced this article was not a good one. We were nine minutes into a quarterly review with a B2B client, the organic sessions chart was down and to the right, and I was explaining the decline. Then their head of sales said something that made the rest of the deck useless: this had been their strongest quarter for inbound enquiries in two years, and more of those enquiries were arriving already knowing what the company did.

Both facts were true. Sessions were down by roughly a fifth. Enquiries were up. And our reporting had no way of holding those two statements in the same frame, because it measured one of them and was structurally blind to the other.

That gap is what this article is about. Not the definition of zero-click search, which you can get from a dozen articles and from the AI Overview that now sits above all of them. The measurement problem: when the click stops happening, what actually happens to the demand, where does it resurface, and how do you show that to somebody who is being asked to justify a budget against a falling line on a chart.

The Short Answer

Zero-click search does not remove demand, it relocates it out of instrumented paths and into uninstrumented ones. To measure it, track five signals together rather than any one alone: query-level impressions holding flat while click-through rate falls in Search Console, branded search volume, direct and unattributed traffic arriving on deep commercial pages, self-reported attribution on your enquiry form, and a manual AI citation panel. No single signal is conclusive. Read together across a full quarter, they reliably separate a brand being cited without clicks from a brand losing relevance.

About the numbers in this article

The observations here come from client accounts we run, aggregated and rounded. They are not a controlled study, there is no holdout group, and the sample is our portfolio rather than a representative slice of the market. Where a figure comes from published research, it is attributed inline. Where it is ours, it is stated as ours and given as a range rather than a decimal.

We have deliberately not put a currency value on an impression. Several vendors will do that for you. The arithmetic is not defensible and it collapses the first time a finance team asks how the join was made.

Why the Traffic Number and the Business Number Came Apart

A click is two things at once, and we habitually forget the second one. It is a visit, and it is a measurement event. The entire analytics stack most marketing teams run is built on the assumption that these are the same thing, because for twenty years they effectively were. If someone encountered your brand through search, they clicked, and the click wrote a row.

An answer served on the results page breaks that assumption cleanly. The encounter still happens. Your content may well be the thing being quoted. But no row gets written, and every downstream system, from the sessions chart to the last-click attribution model to the channel dashboard the CFO looks at, treats the encounter as though it never occurred.

This is not a subtle distortion. It is a systematic one, and it runs in one direction. Every unclicked encounter is invisible, so a channel undergoing zero-click absorption will always look worse in reporting than it is performing in reality. The size of the error grows with how well your content is being used.

Published research gives a sense of scale. Bain's 2025 study found roughly 80 percent of consumers now rely on zero-click results in at least 40 percent of their searches. Estimates of what share of all searches end without any click vary from the high fifties upward depending on methodology, which is a wide enough band that you should not anchor on any specific number. What matters is not the market average but your own rate, and that you can approximate from your own data.

The shape above is the one to look for, and the caveat in the footer of it matters more than the curves. We have watched teams reach for the zero-click explanation to cover a decline whose impressions were falling too, which is a ranking problem wearing a fashionable costume. If you are not certain which you have, work through the Search Console traffic drop decision tree before you go any further with this article. Everything here assumes you have already confirmed the signature.

The Four Places the Demand Actually Goes

If the encounter happened and the click did not, the effect has to surface somewhere. In our accounts it consistently surfaces in four places, in rough order of how quickly it shows up.

Branded search. Someone reads your name in an answer, does nothing, and two weeks later types your company name into Google. This is the highest-volume destination and the easiest to track, because branded query impressions and clicks are sitting in Search Console already. It is also the most commonly ignored, because most SEO reporting filters branded terms out as noise.

Direct and unattributed traffic on deep pages. A remembered brand does not usually arrive at your homepage. It arrives at the service page it heard about, often by typing a query and clicking through without ever registering as a referral, or through a browser that stripped the referrer. Direct traffic to a homepage means very little. Direct traffic landing on a specific commercial page is a strong tell.

Self-reported attribution. The single most useful measurement change we made in the last two years was adding one optional field to the enquiry form. Not a dropdown of channels, which people answer badly, but a free-text box asking where they first came across us. Free text is harder to aggregate and enormously more honest. It is the only instrument that catches a prospect who read about you inside an AI answer and never visited the site at all.

Sales conversation content. The softest signal and the one your sales team already has. When prospects arrive on the first call already using your vocabulary, referencing a framework you published, or asking about a distinction you drew in an article they cannot name, your content reached them. There is no way to log this in GA4. There is a way to log it in a CRM field, and it is worth the argument with the sales team to get one added.

None of these four is a clean substitute for the click. Together they describe the shape of what the click used to represent, which is the most that is honestly available.

The Five-Signal Measurement Framework

This is the part that survived contact with actual client reporting. Five signals, in the order we check them.

Signal 1: The divergence check in Search Console

This is the detection step and it answers one question only: is something intercepting the click.

Open Search Console, set a comparison of the last three months against the same period a year earlier, and look at the Queries tab rather than Pages. You are looking for queries where impressions are flat or up and click-through rate is meaningfully down, with average position roughly unchanged. Position stability is what makes this diagnostic. A click-through decline with a position decline is just a ranking drop.

Then segment those queries into definitional and commercial. Definitional queries are the ones a user could have satisfied with a paragraph, which are the ones an Overview absorbs. Commercial queries are the ones where the user still needs to evaluate, compare or buy. Absorption concentrated in the first group is the expected pattern. Absorption in the second group is a genuine problem and needs different work entirely, which we have written about in the context of ecommerce rankings that stopped converting to visits.

Across the accounts we run, the pattern that recurs is a click-through decline on definitional query sets in the range of a third to a half over a year, with commercial query sets moving far less. That gap between the two segments is the useful number, not the aggregate.

Signal 2: Branded search as the demand proxy

Filter Search Console to queries containing your brand name and its common misspellings, and chart impressions and clicks over the same period. If unbranded clicks are falling while branded impressions climb, you are watching demand get displaced rather than destroyed.

Two cautions. First, branded search is contaminated by everything else you do, so a PR push or a campaign will move it and you need to know your own calendar before reading the line. Second, brand searches from existing customers looking for a login page are not demand. Exclude navigational branded queries where you can identify them, because they inflate the number in the direction you want it inflated, which is exactly when to be careful.

This signal pairs naturally with owning what a searcher then finds, which is a separate discipline covered in brand SERP defense. Driving branded search to a results page you do not control converts a win into a handover.

Signal 3: Direct traffic to deep commercial pages

In GA4, build an exploration with landing page as the dimension, filtered to the Direct channel, excluding the homepage and any page reachable from an email or ad. Chart it monthly.

What you want to see, if the zero-click story is true, is direct sessions rising on service and solution pages specifically. That is the footprint of someone who learned your name somewhere they did not click and came looking for a specific thing. If direct traffic is rising only on the homepage, be sceptical, because homepage direct traffic is where mis-tagged campaigns and internal traffic accumulate. If you have not audited your GA4 setup for that kind of contamination recently, the GA4 insights most teams are missing covers the usual culprits, and internal site search data will tell you what those arrivals were actually looking for once they landed.

Signal 4: Self-reported attribution on the enquiry form

Add one optional free-text field: "Where did you first hear about us?" Do not make it a dropdown. Do not make it required.

Then read the answers monthly rather than aggregating them into a pie chart. The value is in the specifics. Answers naming an AI assistant, or naming an article the visitor never landed on according to your analytics, or naming a comparison they saw somewhere they cannot recall, are all direct evidence of an unclicked encounter converting.

Across our accounts the share of enquiries naming a source that does not appear anywhere in that contact's session history has been consistently large enough that we now treat last-click attribution as actively misleading for content. If you want the broader argument for why single-touch models misread this, marketing attribution models explained covers the underlying issue, of which zero-click is the newest and sharpest instance.

This is the only one of the five signals that produces evidence rather than inference. It is worth more than the other four combined, and it costs one form field.

Signal 5: The citation panel

The four signals above tell you demand is being intercepted and reappearing. None of them tell you whether the interception is quoting you or quoting a competitor. That distinction decides what you do next, and the only way to establish it is to look.

Build a fixed list of thirty to fifty queries that matter commercially. Check them on a schedule, monthly, in a clean browser session, from the locations you actually sell into. Record three things per query: whether an AI answer appeared, whether you were cited, and who else was. Keep the list fixed so the panel is comparable month over month, which is the entire point and the thing most teams get wrong by refreshing the query list each time.

This is manual and it is tedious and there is no complete tool substitute yet, whatever the tool vendors say, because answers vary by location, session and time. The mechanics of running one at scale, including where the tooling genuinely does help, are in AI citation tracking. If the panel shows you absent from answers your competitors appear in, you do not have a measurement problem. You have a visibility problem, and the fix is answer engine optimisation work rather than better reporting.

The Case That Gets Misreported Most Often

The bottom-right branch of that framework deserves its own section, because it is where the comfortable story fails and almost nobody says so.

You can be cited frequently and get nothing from it. Citation is necessary for the zero-click mechanism to pay off, but it is not sufficient. What converts an unclicked encounter into a later enquiry is that the reader retained your name and associated it with something. An answer that quotes your sentence without your brand registering does not do that. Neither does a citation in an answer that is not being read by anyone with buying intent.

The diagnostic is signal 2. If you are being cited consistently and branded search is flat over two quarters, the citations are not landing. In our experience that usually traces to one of three causes: the queries you are being cited on are not queries your buyers ask, your brand name appears as a link label rather than inside the substance of the answer, or the claim being quoted is generic enough that no reader would have a reason to remember its source.

The third is the most common and the most fixable, and it connects directly to what makes content quotable in the first place. Pages that get cited with attribution that sticks tend to carry a specific claim that could only have come from you, which is the argument we made in the content formats LLMs cite and saw play out across fifty D2C brands we audited for AI visibility. Generic advice gets quoted anonymously. A number you measured gets quoted with your name attached.

What We Changed in Client Reporting

The framework above is diagnosis. This is what we actually put in front of clients afterwards, because the diagnosis is useless if the reporting still contradicts it.

We split the dashboard into three layers and refuse to blend them.

Layer one, visibility. Impressions, query coverage, citation panel results, branded search volume. Labelled explicitly as leading indicators. Nobody is judged on this layer.

Layer two, traffic. Sessions, click-through rate, landing page performance, including the decline, shown in full and segmented by definitional versus commercial queries so the audience can see where the loss actually came from. We show the bad number. Hiding it is what destroys the credibility of layer one.

Layer three, outcomes. Enquiries, qualified pipeline, revenue, and the self-reported attribution field read as text. This is the only layer anyone is judged against.

The argument the deck makes is that layers one and three moved together while layer two moved against them. That argument is credible precisely because layer two is shown honestly. A deck that only shows rising impressions next to rising revenue and quietly omits the traffic decline is not a measurement framework, it is a defence, and every experienced finance person recognises the shape of it.

This restructuring is a continuation of a broader cleanup we wrote about in the metrics we stopped reporting to clients, and it rests on the same principle that blog traffic is a poor proxy for content value. Zero-click did not create that problem. It made it impossible to keep ignoring.

When the Traffic Drop Is Just a Traffic Drop

The uncomfortable obligation of publishing a framework like this is saying clearly when it does not apply, because a measurement story that explains away every decline is not a measurement story.

Four conditions, any of which should stop you from reaching for the zero-click explanation.

Impressions are falling too. Then your result is appearing less often, which is a ranking or indexing issue. No amount of attribution reasoning fixes a page Google has stopped surfacing.

The decline is concentrated on commercial queries. Bottom-of-funnel queries still send clicks, because a buyer comparing vendors cannot finish that job inside an answer box. If those are what dropped, you are losing to competitors, not to Google.

Enquiries fell in step with sessions. The entire premise here is a divergence between traffic and outcomes. If both fell, there is no divergence, and the framework has nothing to explain. You have a demand problem.

Two clean quarters have passed and branded search is flat. This is the honest failure condition. If citation presence is up, sessions are down, and no compensating signal has moved across six months, the reassuring interpretation has been tested and did not hold.

That last one is the test we impose on ourselves before we use any of this in a client conversation, and we have had it come back negative. When it does, the correct response is to say the content is not working, not to find a fifth signal that flatters it. If the decline turns out to be structural rather than perceptual, an SEO audit or a content decay review is a more useful next step than a better dashboard.

What We Do Differently Now

Measurement changes are only worth making if they change decisions. Four things changed on our side.

We stopped cutting content on session data alone. The old rule was that a post generating negligible sessions after twelve months came out. Under that rule we would have removed several pages that turned out to be among the most frequently cited in our panel. The rule now requires a citation check before removal, which is slower and has repeatedly saved pages that were earning their place invisibly.

We write for extraction on definitional pages and for depth on commercial ones. A definitional page's job is now to be quoted cleanly, which means a self-contained answer block near the top and a specific claim inside it. A commercial page's job is still to convert a visitor, because those clicks survive. Treating both page types the same was producing content that did neither job well, a split we worked through in more detail when running SEO, AEO and GEO for a single client.

We front-load the claim. If a page's distinctive contribution appears in paragraph nine, it will not be lifted. The mechanics of this are the same ones that govern ranking in AI Overviews, and they are structural rather than editorial.

We ask about zero-click in the sales process. One question on the first call about where the prospect first encountered us, logged in the CRM. It is soft data. It is also the only place the fourth destination shows up at all.

The Mistakes We Made First

Worth stating, because each cost us a quarter.

We used impressions as a success metric on their own. Impressions rise for reasons that have nothing to do with performance, including Google simply showing more results per query, and a deck built on impressions alone falls apart under one good question.

We tried to model a value per impression. The number was arbitrary, everyone downstream knew it was arbitrary, and it damaged the credibility of the honest parts of the report.

We refreshed the citation panel query list every month, which meant we had twelve months of data and no trend, because nothing was comparable to anything else.

We assumed citation equalled benefit for most of a year before checking branded search against it, and were wrong for at least one client where the citations were real and the brand recall was not.

What We Still Cannot Do

There is no join key between an impression and an enquiry. Everything above is correlation held across enough accounts and enough quarters that we act on it, and none of it is proof for a single case. We cannot tell you what one AI citation is worth. We cannot separate zero-click effects from the general rise in brand awareness that any active marketing programme produces. And the manual citation panel, which is the most valuable of the five signals, is also the least scalable part of the whole framework.

Anyone offering you a clean attributed number for zero-click revenue is modelling and presenting it as measuring. The distinction matters, and holding it is what makes the rest of the argument trustworthy.

Where to Start

If you want to run this on your own account, the order matters and the first two steps take an afternoon.

Start with signal 1, because it is free and it tells you whether the rest is relevant. If impressions are not stable, stop and diagnose a ranking problem instead. Then add signal 4, the form field, because it is the only one producing evidence rather than inference and it starts accumulating from the day you ship it. Signals 2 and 3 need a quarter of data before they say anything. Signal 5 is the one that requires real discipline, and it is the one that will change what you do next.

For the wider context on how the search landscape produced this problem, generative search explained covers the mechanics, SEO vs AEO vs GEO covers how the three disciplines divide, and our AI search statistics page tracks the published numbers with sources attached. If the terminology in any of this is unfamiliar, the digital marketing glossary defines it.

If your reporting is currently telling a story your sales team disagrees with, that disagreement is data. It usually means the instrumentation stopped matching the way people actually find you, and the fix is a measurement change before it is a content change.

If you would rather have the five signals run for you across your full commercial query set, that is the work our AI SEO team does, alongside the technical SEO and SEO programmes the measurement sits on top of. Tell us what your numbers are doing and we will tell you which of the four cases you are in.