You pull the date range in Search Console back to the last three months and compare it with the three months before that. There is an up arrow in the impressions column and a down arrow in the clicks column. Average click-through rate has landed below the previous period. The first sentence of the meeting is already written: "SEO is broken, traffic is falling."
Maybe nothing is broken. Your page still appears for the same queries — it appears more often, in fact. The user no longer needs to click, because the answer has already been given at the top of the results page or inside a chat window. What is falling in that table is not demand; it is the way demand reaches you. Your measurement setup was built for the old way, which is why it cannot see the new one at all.
What follows is not a conceptual argument. It is a measurement routine you can run with the tools already on your desk: which screen to look at, how to read two numbers side by side, which fluctuations mean nothing. The last section says where the routine jams — because it does jam.
What an impression without a click is worth
A click is the most easily measured event showing that a user reached you. The claim that it is the most important one is a separate matter, and it is not true. An impression that never becomes a click can still leave three things behind:
- Name recall. The user reads the answer, sees your brand name inside it, and does nothing at that moment. A week later, when the need turns concrete, they type your name into the search box rather than the name of the service.
- A place on the shortlist. When an assistant offers three options, the user usually does not step outside those three. Getting onto the list produces no click by itself, but it moves you onto the short list.
- Category trust. "This company has published a calculator on that subject" is information that prepares the ground for the next contact even when no click happens.
All three share one property, and that property is exactly what makes measurement difficult: the signal is delayed. The effect appears not on the day of the impression but days or weeks later, in a different channel. Counting zero clicks as zero value is a mistake that comes from never putting that delay into the measurement window.
The delayed signal has a bad side as well. It is hard to manage, hard to prove, and in bad-faith reporting it can be used to explain away anything. "There are no clicks, but the brand is known" is an excuse when there is no number under it. Every section below is about producing that number.
The impressions-to-clicks gap in Search Console
In Search Console's Performance report, tick all four boxes at the same time: clicks, impressions, average CTR, average position. Set the date range to the last 6 months and read the chart for one thing only — is the distance between the impressions curve and the clicks curve widening over time?
A widening gap proves nothing on its own. The same picture comes out of all three of the following situations:
| What you see | Possible cause | How to tell it apart |
|---|---|---|
| Impressions rising, clicks flat | You are appearing for new queries and the answer is being given on the results screen | In the query breakdown, check whether the new queries carry informational or purchase intent |
| Impressions rising, clicks falling | Average position may have slipped backwards | Check the position column separately for the same query |
| Impressions flat, clicks falling | The title or description changed, or a competitor's snippet got stronger | Compare page by page in the page breakdown |
For a sound reading, break the gap down at query level. Move to the Queries tab in the Performance report and sort CTR in ascending order. The queries that stay at the top — high impressions, low CTR — are your zero-click cluster. These are usually questions with a one-sentence answer: "what is", "how long does it take", "how much does it cost". Having that answer delivered on the results screen is expected behaviour.
The interpretation error starts here: treating a CTR drop on its own as failure. CTR is a ratio, and when its denominator grows it falls even if the numerator stays put. In a period when you begin appearing for new queries, a falling CTR is most often the arithmetic consequence of visibility widening. The thing to look at is not the ratio itself but whether the absolute number of clicks has fallen. If the click count is flat while impressions rise, nothing has been lost; reach has expanded. If the click count is falling too, there is a real problem, and that is the point at which you look at position and snippet. We dealt separately with how answer boxes are changing the results screen in how Google AI Overviews affect businesses.
Brand search and direct traffic: reading the delayed signal
The place where an unclicked impression pays off is not the query that produced it. It is two separate line items.
The first is brand queries. In Search Console, type your brand name into the query filter — try the common misspellings separately as well. Record the impression and click totals of that cluster monthly. A brand query is the cleanest indicator that the user already knows you; nobody types the name of a company they have never heard of into a search box.
The second is direct traffic. In analytics, the "Direct" bucket holds both the people who genuinely typed the address by hand and the sources that carry no referrer information. That makes it insufficient as evidence on its own; but once you have split traffic coming from AI assistants into its own channel, what remains in Direct becomes more readable. How to set up that channel split is laid out step by step in track AI traffic in GA4.
Be clear about the comparison window. Two rules do the work:
- Look monthly, not weekly. Brand search is low in volume; on a weekly chart it drowns completely in noise.
- Compare yourself with yourself, not with the sector. The benchmark is your own figure from last month and from the same month last year. Where there is seasonality, the year-on-year comparison is mandatory.
The lag has to be built in too. The distance between the moment of information seeking and the moment of need varies by sector: hours for an emergency repair service, weeks for a treatment or a consulting decision. That is why "we published content this month and brand search did not rise this month" is not a conclusion. If you know the length of your own sales cycle, shift the lag window to match it.
Why hiding the answer does not work
The most common reflex against zero clicks is this: leave the answer half finished and say "see our site for details". In today's search environment that move has no return. If the model cannot get the answer from you, it takes the same answer from another source that does give it, and shows that source. The only thing you gain by withholding the answer is not being the party that gets cited.
The construction that works is the opposite: give the answer plainly in the first paragraph and put the reason to click in what comes after it. The reason to click has to be something that cannot be copied:
- A tool or a calculation. A calculator the user feeds their own figures into cannot be quoted as a piece of text.
- A template or a checklist. A structure that can be downloaded, or ticked off item by item on the page.
- Local information specific to you. Your price band, your working hours, which district you travel to, which institution you have an agreement with.
- Current data. A table that shows its date and is updated regularly. A model is cautious about giving a current figure and generally points to the source instead.
The practical consequence of this split: the "answering" part of a page and the "click-earning" part do different jobs, and the two can sit side by side on the same page. The citation logic described in how to get cited by Perplexity arrives at the same place — to be cited, you have to give the answer.
Page construction that turns clickless impressions into conversion
If the user is never going to arrive on your page, you gain as much as the trace left inside the answer is worth. Three things work directly towards that.
The brand name appearing inside the answer. In the text, use the full written form of the brand name at intervals instead of "we", "our firm", "our team". When a model tears a paragraph out of its context, the word "we" means nothing; a brand name, on the other hand, travels into the answer. Do not overdo it — a name crammed into every sentence tires the reader and does not read naturally.
A one-sentence positioning line. Say who you are on the page in a sentence that stands up on its own: what you do, for whom, where. That sentence has to be built so that it can be lifted into an answer exactly as it is. "Quality service with years of experience" is not positioning; it answers no question at all.
Contact details being machine readable. The phone number and the address must not be buried inside an image, written in afterwards with JavaScript, or sitting behind an email obfuscation layer. They have to be present as plain text in the page's source HTML and declared with structured markup as well. In a clickless world, if the contact detail can get inside the answer, an impression can turn straight into a phone call; if it cannot, the impression ends there.
Who zero-click is a real threat to
This is not a problem of equal weight for everybody. The dividing line is the revenue model:
| Type of business | Where the revenue comes from | Effect of zero-click |
|---|---|---|
| Information publisher, content site | Page views, advertising, subscriptions | Direct revenue loss; the click is the product itself |
| Local service business | Appointment, phone call, quote | Channel shift; the click was always an intermediate step |
| E-commerce | Cart and checkout | Mixed; loss on informational queries, clicks continue on product queries |
| B2B consulting | Long-cycle conversations | Usually neutral; getting on the shortlist matters more than the click |
For a local clinic, lawyer or estate agency the click was never the goal; the goal was the phone ringing. If the user sees you in an assistant's answer and calls you directly, never having visited your site is not a loss but a shortened path. In these businesses the real risk is not zero clicks — it is not appearing inside the answer at all. You can see how this flow is built for your own sector, with concrete examples, on the use cases page.
On the information publisher side the problem is real, and it is not solved by measurement; it is solved by the business model. The discussion there is not about click measurement but about whether a revenue line exists that does not depend on clicks. We covered separately where the paid side moves as clicks decline in our article on ads in AI Overviews and AI Mode.
The monthly reporting routine
Look at fewer than three numbers and you will be wrong; look at ten and you will not be able to follow any of them. The trio to be read together each month is this:
- Total impressions (Search Console, non-brand queries) — tells you whether visibility is widening.
- Brand query volume (Search Console, brand filter) — tells you whether that visibility is sticking in memory.
- Qualified contact count (form, phone, WhatsApp, appointment) — tells you whether any business comes out at the end of all of it.
These three are read together. If impressions rise while brand search and contacts stay flat, the visibility is not working. If impressions stay flat while brand search and contacts rise, another channel is doing the work. If all three rise together, the direction is right. Click count and CTR are supporting columns in this table, not the headline. A paid channel enters this table as its own row, and ad management reporting keeps that row separate. For the full list of metrics on the visibility side, see how to measure AI visibility.
The noise threshold has to be set in advance too. In low-volume data, single-digit movements say nothing: fifteen brand searches becoming twenty-one is a fluctuation, not a trend. As a practical rule, take the average of three consecutive months and do not hang a decision on a single month spiking. A campaign, a season, a piece of press coverage, a competitor shutting down — all of these produce a one-month jump, and none of them is the result of your work.
The limit of this measurement
To be honest about it: no tool today can show the chain "the user read you in an assistant's answer, searched your name two weeks later, then telephoned" from end to end. The impression on the chat screen does not happen on property you own, so the measurement does not start on your property either. You can only establish the link with indirect evidence: curves that move at the same time, a "how did you hear about us" field on the form, a single question asked on the phone.
These are weak forms of evidence, and saying that they are weak is better than behaving as though they were strong. The most solid thing you can do is to look at the joint movement of trends rather than at a single metric, and to write the margin of error openly when you make a decision. If there is a tool or a report claiming to show you an end-to-end attribution chain, ask what data it rests on; the answer is most often a modelled estimate.
Explaining that uncertainty to a buying committee is a separate job; our B2B and enterprise solution page covers how that conversation is set up.
Frequently Asked Questions
My click-through rate fell but my impressions rose, is that a bad sign?
Not a bad sign on its own. Click-through rate is a ratio, and when impressions, its denominator, grow, the ratio falls even if the click count stays the same. What you should look at is the absolute number of clicks: if that number runs flat while impressions rise, your visibility has widened. If the click count is falling as well, that is when you need to look at your average position and your page titles.
Is zero-click equally dangerous for every business?
No, it changes completely with the revenue model. For a content publisher whose revenue comes from page views or advertising, the click is the product itself and losing it is a direct revenue loss. For a local service business working on appointments, phone calls or quotes, the click was never the goal, only an intermediate step; if the user sees you in the answer and calls directly, the path has simply got shorter.
Does hiding the answer on the page to force a click work?
It does not. If you do not give the answer, the model finds another source that does give it and shows that source; you are not the party that gets cited. The right construction is to give the answer plainly in the first paragraph and to tie the reason to click to something that cannot be copied as text: a calculator, a downloadable template, price and local information specific to you, or a regularly updated data table.
Over what time range should I measure the rise in brand search?
Look monthly, not weekly. Brand search is low in volume at most businesses and disappears completely into noise on a weekly chart. Make the comparison against your own past figures: the previous month, and the same month last year if there is seasonality. Also take the length of your own sales cycle into account, because weeks can pass between the information search and the moment of need.
Which tool can track a customer who read about me in an assistant and called later?
There is no tool today that shows this chain end to end. The impression on the chat screen does not take place on property you own, so the measurement does not start there either. What can be done is to collect indirect evidence: keep a "how did you hear about us" field on the form, ask a single question on the phone, and watch whether impressions, brand search and contact counts move together. Admitting that this evidence is weak is healthier than reporting it as though it were strong.