AccuRanker Analysed 1.3 Billion SERPs. What Does That Really Tell Us About CTR?

AccuRanker recently published a whitepaper based on 1.3 billion US desktop and mobile search results. The findings about rankings, click-through rates, search intent, pixel position and AI Overviews are interesting. There is quite a lot in it that SEO people should pay attention to.

There is also a fairly obvious commercial angle. The company sells the dynamic CTR model used in the analysis. That is fine. Research does not become wrong because it supports the product of the company that produced it. But it does mean I want to keep three things apart: what the report actually observed, what the model estimated and what we can sensibly conclude from either.

So I read the CTR report, then compared its claims with published research based on observed search behaviour, aggregated Search Console data and academic studies. I also looked at what the findings might mean for SEO and, where it is actually relevant, GEO and AI search.

My short version is this: the report identifies a real problem. Rankings do not predict clicks very well without context. But a more complicated proprietary model is still a model. It does not turn an estimate into a reliable traffic forecast simply because more factors went into it.

What the report found

The report uses Google US search results collected in January 2025. It covers desktop and mobile and examines more than 120 factors, including rank, pixel position, search intent, title characteristics and SERP features.

Some of its headline findings are striking:

• Average position-one CTR was approximately 26% on desktop and 18.5% on mobile, but individual estimates ranged from almost nothing to around 75%.
• Mobile SERPs containing an AI Overview received roughly half the organic traffic of those without one.
• Navigational searches produced position-one CTR above 40%, while transactional position-one CTR was around 16% on desktop and below 11% on mobile.
• A mobile featured snippet received an average CTR of approximately 12%, considerably below position one on an informational SERP without one.
• Pixel depth had a strong relationship with CTR, although the report found no sudden cliff at a supposedly universal 1,080-pixel “fold”.
• Question-led searches tended to receive fewer external clicks, while terms such as “buy”, “sale” and “login” produced more.
• Shopping results reduced organic CTR, particularly on mobile.

The common thread is sensible. A position number tells you almost nothing about the result page around it. Position one below an AI Overview, four adverts and a Shopping carousel is plainly not the same as position one on a fairly clean branded search.

This should not be controversial, but SEO forecasts still regularly multiply search volume by a generic position-based CTR percentage and present the result as future traffic. It looks tidy in a spreadsheet. That does not make it likely to happen.

The 1.3 billion figure needs a closer look

The size of the dataset sounds impressive, and it is. But the precise wording matters.

The report is based on 1.3 billion SERPs. It is not based on 1.3 billion observed clicks. The CTR figures come from a proprietary AI-based model, and that distinction matters.

The model apparently considers more than 120 parameters. What the whitepaper does not give us is the information we would need to test it independently: the click data used for training and validation, error rates, calibration, confidence intervals, useful segment sizes or performance on unseen data.

The company positions the model as a way to estimate click potential and forecast business impact. Including more than 120 factors should make it more useful than a generic static CTR curve. However, the public report does not give us enough information to judge how accurate those estimates are at individual-keyword level. I would therefore treat them as better-informed forecasts rather than precise predictions.

A large collection of SERPs can show us how often features appear and where results sit on the page. It cannot, by itself, tell us where users clicked, so the quality of the estimate also depends on the behavioural data and validation behind the model.

Those details may understandably remain proprietary. The practical point for an external reader is simply that the output is a useful estimate, not an independently verifiable benchmark.

The main conclusion holds up: rank alone is not enough

Independent research supports the central argument.

A 2023 study of 67,000 keywords across more than 40 US ecommerce domains used 24 million impressions and six million clicks to examine the effect of 24 SERP features. The researchers found that those features added predictive information beyond ranking position and could increase or reduce CTR depending on the feature, the rank and whether the website was included in it. Position remained highly important, but it was not the whole explanation. The study is appropriately called Beyond Rankings.

Another peer-reviewed study used 31.6 million impressions and 416,386 clicks to estimate device-specific organic CTR. It placed the average position-one CTR at 9.28%, followed by 5.82% and 3.11% for positions two and three. That is a long way from the report’s 26% desktop and 18.5% mobile estimates. The researchers also found meaningful differences between devices.

That does not prove one study right and the other wrong. The data, samples and methods are different. It does show why a universal CTR curve is a poor base for a precise forecast.

The average changes when the mixture of branded and non-branded searches changes. It changes with device, country, industry, intent, SERP layout and the websites represented in the dataset. It may also change depending on how average position and impressions are calculated.

There probably is no single position-one CTR waiting to be discovered. There are just averages from different collections of searches.

Pixel position is useful context, not a replacement for rank

The report argues that pixel position can tell us more about CTR than rank. The useful part is straightforward: organic rank gives us the order of the organic listings. It does not tell us how far down the page somebody has to go before seeing them.

Modern research makes the problem visible. A 2026 analysis of 46,000 US and UK search terms found that the median desktop position-one organic result began approximately 635 pixels down the page. Only 57% of position-one organic results appeared above a typical desktop fold; on smartphones, the figure was about 40%. The study measured SERPs in pixels rather than relying only on rank.

That is useful information for rank tracking and visibility analysis. It does not mean rank has become irrelevant or that a position-two result near the top will generally outperform position one further down.

The report itself still says rank matters. Its pixel-depth graph also mixes results with different ranks, intents and SERP features, so it cannot isolate what happens when an otherwise identical listing simply moves further down the page.

Nor should we expect CTR to collapse at exactly 1,080 pixels. The fold depends on the device, viewport, browser interface, zoom level and orientation. There is no universal line at which every searcher suddenly loses interest.

Pixel position improves the description of visibility. It does not provide a new law of user behaviour.

The evidence that AI Overviews reduce clicks is now difficult to dismiss

The report finds a large effect from AI Overviews across the top organic positions, especially on mobile. Other studies, using quite different methods, point in the same direction.

Pew Research Center observed 68,879 Google searches made by a representative panel of 900 US adults. Users clicked a traditional result in 8% of visits when an AI Overview appeared, compared with 15% when one did not. Only 1% clicked a source inside the overview. Users were also more likely to end their browsing session after seeing an AI Overview.

Ahrefs took a different approach. It compared 150,000 keywords that triggered AI Overviews with 150,000 informational keywords that did not, using aggregated Google Search Console data. Its February 2026 update estimated that an AI Overview was associated with a 58% reduction in position-one CTR. The estimated effect became smaller further down the top ten, but remained negative.

Seer Interactive analysed 25.1 million organic impressions across 3,119 informational and educational queries from 42 organisations. Organic CTR for its AI Overview query set fell from 1.76% in June 2024 to 0.61% in September 2025, a 61% decline. Its longer-term analysis also found that queries without AI Overviews were losing CTR, although more slowly.

There is now experimental evidence too. A 2026 working paper based on a field experiment found that, when AI Overviews appeared, they reduced outbound organic clicks by 39.8% and increased zero-click searches. As a working paper, it should not be treated as the final word, but its experimental design helps address some of the limitations of observational studies.

The numbers are different, but they all point the same way.

But AI Overviews are not the whole explanation

There is a complication in the Seer research which should not be brushed aside. Queries that displayed AI Overviews in September 2025 already had lower CTR earlier in the data period, before those overviews appeared.

That makes sense. Google is more likely to generate an answer for informational searches that could already be resolved through a featured snippet, knowledge panel or another instant-answer feature. The searches most suited to AI Overviews were not necessarily high-click opportunities waiting for AI to ruin them.

AI Overviews can reduce clicks while also being selected for query types that were already less likely to produce clicks. Both can be true.

This is why a statement such as “AI Overviews have cut organic traffic by 58%” needs some care. Ahrefs estimated the effect on the top-ranking page within selected keyword groups. Seer’s 61% was a change over time among informational and educational queries. Pew measured what its user panel did next. These are not three versions of the same measurement.

Google offers a counterargument. In August 2025, it said that total organic click volume to websites had remained broadly stable year on year and that the clicks it sent were of higher quality, meaning users were less likely to return quickly to the results. Google did not publish the underlying data.

Its claim is not necessarily incompatible with falling CTR on AI Overview queries. If people conduct more searches, total click volume can remain stable while the percentage of searches producing a click falls. Aggregate traffic can also stay level while being redistributed between websites and query types.

Google may be sending a similar number of clicks overall. That will be limited comfort if your website is the one losing them.

Intent matters, but tidy labels can hide messy behaviour

The findings about intent look plausible. Navigational searches concentrate clicks on the intended brand. Informational searches are easier to answer on the results page. Transactional results are full of adverts, Shopping features, marketplaces and comparison options fighting for attention.

The report estimates position-one navigational CTR above 40%, compared with around 16% for transactional searches on desktop and below 11% on mobile. It also suggests that lower organic positions capture a meaningful share of transactional clicks.

I would use the direction. I would not lift the percentages into a client forecast as though they were observed facts.

The intent categories come from another AI model trained on labelled and unlabelled data. The company reports more than 90% agreement with human experts, but does not publish the validation sample or explain how mixed intent is handled.

There is also an ambiguity in the lower-ranking transactional finding. The report’s prose refers to positions six and higher, while its chart is labelled positions 6–100. That is not sufficient evidence for the more attractive claim that 23% of transactional clicks go specifically to positions six to ten.

The practical lesson is not that we should all chase position seven because comparison shoppers apparently love it. It is to look at why somebody is searching, what Google puts around the organic results and whether that query represents a useful business opportunity in the first place.

Featured snippets, Shopping and question queries are not simple wins or losses

The model estimates an average mobile CTR of 12% for a featured snippet, compared with more than 25% for position one on an informational SERP without one. The explanation is believable: sometimes the snippet answers the question and there is no reason to click.

That does not make featured snippets undesirable.

Owning a featured snippet may produce fewer clicks than owning position one on a clean SERP. It may still be far better than remaining in an ordinary position four while a competitor owns the answer above you. The useful comparison is not with an imaginary perfect SERP. It is with the realistic alternative available for that query.

The same applies to Shopping results and question terms. Their presence can reduce total organic CTR, particularly on mobile, but a query containing “what” does not become commercially worthless because its average click rate is lower. It may introduce a buyer to a category, resolve an objection or influence a decision completed later.

Click volume matters. So does what the click, impression or answer contributes to the business.

What does this mean for GEO?

AI Overviews sit inside Google Search, so they are clearly an SEO issue. They also belong in the wider AI-search discussion which we have, for better or worse, started calling GEO.

The obvious response is to pursue citations inside the overview. Seer found that brands cited in an AI Overview had a 35% higher organic CTR than brands that were not cited: 0.70% compared with 0.52% in its Q3 2025 comparison.

That is not the same as capturing 35% of the remaining clicks. Nor does it prove that the citation caused the improvement. Stronger brands may be more likely both to attract clicks and to be cited.

More importantly, a citation does not restore the old traffic model. It may provide visibility within an environment where fewer people leave Google, but visibility, attribution and commercial benefit are different outcomes.

A source can be cited without its brand being recommended. Its information can shape an answer that favours a competitor. A mention can be technically accurate but commercially irrelevant. I have written separately about why businesses should stop counting citations and examine whether AI actually understood the brand.

So GEO does not solve the CTR problem. It gives us another set of signals to interpret, and another opportunity to put an impressive-looking number into a dashboard without asking what it means.

If the industry replaces “we rank number one” with “we were cited in 37% of tracked prompts”, it has not improved its strategic thinking. It has changed the screenshot.

A more honest way to forecast SEO opportunity

CTR models are not useless. Businesses need some way to estimate opportunity, compare priorities and make investment decisions before the clicks exist. The problem starts when an estimate is dressed up as precision.

A more defensible forecast starts with the website’s own evidence. Search Console can show how CTR varies across devices, countries, query groups and pages at broadly similar positions. Those data have their own limits, including anonymised queries, average-position complications and changing SERPs, but they describe the website and audience in question rather than somebody else’s average.

External CTR studies and SERP models can add context. I would use them to challenge an assumption, not to make a promise.

For important opportunities, I would consider:

• the site’s historical CTR for comparable queries and pages;
• branded versus non-branded demand;
• device and country differences;
• current AI Overviews, adverts, Shopping, local results and other SERP features;
• where the organic result physically appears;
• whether the query is informational, comparative, navigational or action-led;
• the commercial value and conversion behaviour of the resulting visit;
• how often the result-page layout changes.

The output should normally be a range, not one reassuring number. A conservative case, a plausible case and an optimistic case make the uncertainty visible. If the investment only works in the optimistic version, the decision-maker should know that before approving it.

Search evidence can also do more than estimate channel traffic. It can reveal customer language, objections, comparisons and shifts in demand. That is why I prefer treating search data as market intelligence rather than leaving it trapped inside an SEO forecast.

The report is right about the problem

The report is worth reading. Its most important conclusion is well supported: CTR is shaped by far more than organic rank, and a static curve will often misrepresent the opportunity.

Where I part company with it is the leap from that observation to confidence in a proprietary prediction for every keyword.

Adding intent, pixel position and SERP features should produce a more informed estimate. It does not remove uncertainty about user behaviour, future result pages, demand, conversion or commercial value. A sophisticated model can still offer false confidence if its output is presented without an honest range and an explanation of what remains unknown.

Search results have become more complicated. Our forecasts should become more honest about uncertainty, not just more elaborate.

Rankings still matter. Clicks still matter. AI citations may matter. None of them is a business outcome on its own.

The useful question is not whether a tool can give us a more precise CTR percentage. It is whether the evidence is good enough to support the decision being made.

---