Zero click search: attributing organic traffic when the click never comes

Not a new phenomenon: the scale changed

Anyone looking up the weather, an opening time or a currency conversion never had a reason to open a page. What changed is the scale: the surfaces that end the search on the page have multiplied, and now include synthesised answers to involved questions, not only single data points.

For a publisher the practical problem is not philosophical. It is that the channel keeps producing visibility while the metric it has always been judged by, the visit, stops representing it.

The three statistics that get confused

Before any measurement, one misunderstanding is worth clearing up, because it makes most of what is written on the subject unusable. Three families of numbers circulate, and they get quoted as if they measured the same thing.

Three different denominators. A percentage of searches does not compare with a percentage of tracked keywords, and neither of them says how many people do anything. Yet the three numbers are routinely chained together in one paragraph to build a trend.

There is a second, subtler error, about how surveys are read. A result along the lines of “eighty per cent of users rely on generated answers in at least forty per cent of their searches” is a figure about people, not about searches: it says how many individuals have that habit, not that eighty per cent of searches end that way. The difference is substantial and gets inverted constantly.

What Search Console actually measures

The performance report's four metrics have precise definitions, and almost every misreading comes from never having read them.

• Impression: how often a user saw a link to your site on Google. For standard results it counts when the element appears on the page, regardless of scrolling, but only provided the user did not have to click to see more results. For scrollable widgets such as carousels, the element generally has to be scrolled into view, or expanded by a click.
• Click: how often a user clicked a link on Google that took them to your site.
• CTR: simply clicks divided by impressions. It is not a judgement of quality, it is a ratio between two numbers that can move independently.
• Average position: a relative ranking of your link's position on Google, where 1 is the highest, averaged across every query the property appeared on. Google attaches an explicit warning: it is a complex metric that can be misleading if you do not understand its subtleties.

It is worth understanding why. Under property aggregation the count takes, for each query, only the highest position the site occupies. In the documentation's own example, a query returning the property at positions 2, 4 and 6 counts as 2; a second query returning it at 3, 5 and 9 counts as 3; the average position is 2.5.

An unintuitive consequence follows: in this view, losing the same site's lower placements on a query does not move the average position, because the second place still defines it. Some of the clicks go, though. Under page aggregation the count changes, because each URL is counted on its own and the average shifts.

How generated answers are counted

The documentation is precise on this, and it is rarely quoted. A click on a link leading out of an AI Overview is counted as a click, exactly like any other. For impressions the standard rules apply, and the standard rules already carry an important condition: the impression counts when the element appears on the page, regardless of scrolling, provided the user did not have to click to see more results. A link already visible inside an AI Overview therefore produces an impression like any other; a link folded behind an expansion produces one only if somebody expands it.

On position the documentation is equally precise: an AI Overview “occupies a single position in search results, and all links in the AI Overview are assigned that same position”, while position in AI Mode follows the same methodology as a Google Search results page. Finally, a limit worth remembering when the numbers seem not to add up: Search Console does not include data from experiments in Search Labs.

What Search Console shows of generated answers, and what it does not

Much of what is written on this point aged within a few weeks. The starting point is the AI features page, where Google states that sites appearing in AI features are included in the overall search traffic in Search Console, specifically in the performance report, under the Web search type. For a long time that was the only guidance available, and it is where the idea repeated everywhere came from — that traffic from generated answers was entirely invisible.

That is no longer the case. There is a generative AI performance report, which Google made available to all websites globally from 31 August 2026, and which includes impressions for AI Overview and AI Mode.

It is worth being precise about what it contains, though, because it is a good deal less than the name suggests:

• It exposes impressions only. No clicks, no CTR, no average position.
• There are no queries. The available dimensions are pages, countries, devices and dates.
• AI Overview and AI Mode cannot be told apart: the report has no dimension separating them, so the two surfaces' impressions arrive combined.
• The usual performance-report limits apply, starting with the thousand-row cap.

The result is half an answer, and it matters to know which half is missing. Knowing which pages the site appears on inside generated answers is now possible. Knowing how many of those clicks reached the site is not: the clicks stay mixed inside the “Web” search type along with all the others. Anyone promising to quantify “how much traffic I lost to AI” using Search Console is still estimating, not measuring, and should say so.

On the same AI features page, Google states that clicks from results pages with AI overviews are “higher quality”, and that users are more likely to spend more time on the site. It is worth knowing the claim exists, and equally worth knowing that Google does not publish the data behind it: it is a statement from an interested party, not a verifiable result.

Consistently, Google recommends monitoring conversions and time on site in other tools too, such as Google Analytics. That is sensible advice for a reason that goes beyond AI: Search Console measures what happens on the results page, not what happens after the visit.

Diagnosing “clicks are falling, impressions are not”

It is the pattern that sends most people looking for information about zero click search. It does not have a single cause, and which cause it is changes entirely what to do about it. Here is the procedure, in order.

1. Check you are not reading provisional data. Google warns that the most recent data in the performance report can be preliminary — still being collected, and liable to change in the hours that follow. Always exclude the recent tail before crying collapse.
2. Check which aggregation you are looking at. The chart and the table do not count the same way: the chart aggregates by property, while for pages and search appearance the table aggregates by page. Comparing the chart's total with the table's sum produces discrepancies that mean nothing.
3. Compare two periods at the right granularity. In the advanced filter and comparison Google suggests considering weekly or monthly granularity when comparing two date ranges. On daily data the weekend noise looks like a trend. Only one comparison can run at a time, so set up one scenario at a time.
4. Separate brand queries from the rest. The “Queries containing” and “Queries not containing” filters isolate the company name. Mind the matching rule: it is not case-sensitive, but is otherwise exact, spaces included. So the recurring variants and typos have to be accounted for, or they end up on the wrong side.

At that point the picture reads like this.

The last row deserves attention, because it is the one that prevents hasty decisions. If revenue is not moving while visits fall, the problem might not be a problem.

One caution on method remains: none of these readings proves the cause on its own. A falling CTR at the same position is compatible with an AI Overview appearing above you, but also with a lost featured snippet, with new ads at the top, with a competitor rewriting their title, with seasonality. The table narrows the field of hypotheses; it does not close it.

Brand demand as the counter-check

If the thesis is that visibility produces effects before the visit, there has to be an independent signal confirming it. Brand queries in Search Console are the first. Google Trends is the second.

But you need to know what you are looking at, because Trends data is not volume. Each data point is divided by the total searches of the geography and time range it represents, and the numeric results are then scaled into a 0 to 100 range. The value 100 is the peak of the series you asked for, not a quantity.

Three stated limits that change the interpretation: Trends uses only a sample of Google searches; it shows data only for the more frequent search terms, and reports 0 for terms with low search volume; and it removes repeated searches from the same person over a short period.

For a small brand the consequence is concrete: a flat line at zero does not say nobody is searching for you, it says you are below the threshold. In that case the more useful source remains brand queries in Search Console, which are about your site rather than a market sample.

Reading year on year without fooling yourself

Comparing against the same period the previous year is the most honest way to look at a seasonal channel, and also the easiest way to tell yourself a comfortable story.

Three rules keep the comparison clean. Compare periods of the same length containing the same number of weekends, because a month with five Mondays against one with four produces a gap by itself. Account for movable holidays, Easter above all, which shift whole blocks of demand between March and April. And check that the site's perimeter has not changed in the meantime: pages removed, sections migrated, a domain merged in — these explain gaps that have nothing to do with the SERP.

What this data does not let you conclude

It is worth being explicit, because this is the part that is systematically missing elsewhere.

• You cannot quantify the traffic lost to generated answers: the dedicated report shows impressions, not clicks, and the clicks stay indistinguishable inside the total.
• You cannot infer that you were cited in an AI Overview from the site's overall impressions, because those sum everything: for that question there is the generative AI report.
• You cannot prove that a citation in a generated answer produced a conversion, because there is no identifier between the citation and the visit linking the two.
• You cannot turn the absence of clicks into a positive story without a second signal supporting it: brand demand, conversions, direct enquiries.

What you can do is more modest and more useful. You can establish whether the channel is losing positions or only clicks, whether people are arriving by other routes, and whether the value produced has changed. Those are three questions the data genuinely answers.

Common questions