Research
The disappearing click: what Google's top spot is worth in 2026
Most coverage of AI search argues about whether it will matter. That argument is over, and it was settled by measurement rather than opinion. Three large studies published this year put numbers on what an AI answer costs the page underneath it. Read together they describe something more uncomfortable than a traffic dip: the position you spent a decade earning now pays roughly half of what it used to, and the thing that replaced it cannot be won by ranking.
Key takeaways
- When an AI Overview is present, the top ranking page takes 58% fewer clicks. Ten months earlier the same study measured 34.5%.[1][2]
- The loss runs the length of the page. Position 5 loses 32.6%, position 10 loses 19.4%, so outranking a rival does not recover it.[1]
- 68.01% of US Google searches now end without any click, up from 60.45% in 2024. Only 276 of every 1,000 searches reach the open web.[3]
- Only 37.9% of the URLs Google cites inside its own AI Overviews come from its own top 10. Nearly two thirds sit outside page one.[4]
- The demand did not disappear, it relocated: 51% of B2B buyers now open product research in a chatbot, against 29% a year earlier.[5]
In this article
What the top spot lost#
In April 2025, Ahrefs compared 300,000 keywords, half of which triggered an AI Overview and half of which did not, and found that the presence of an AI Overview correlated with a 34.5% lower clickthrough rate for the top ranking page.[2] The number circulated widely and became the standard citation for the effect. It is now out of date. Ahrefs re ran the study on December 2025 data and published the result in February 2026: the same comparison now shows a 58% lower clickthrough rate for the top ranking page.[1]
Clickthrough loss at position 1 when an AI Overview is present, measured ten months apart on the same 300,000 keyword design. Ahrefs, April 2025 and February 2026.
The detail that matters more than the headline is what happens further down. This is not a penalty applied to the winner while everyone below carries on as before. Position 2 loses 50.8%, position 3 loses 46.4%, position 4 loses 38.8%, position 5 loses 32.6%, and even position 10 loses 19.4%.[1] Every rung of the ladder is shorter than it was.
Two honest qualifications, because this article is only as good as its worst citation. The study measures correlation across a two year gap, December 2023 against December 2025, and the authors say so plainly. Some of the decline belongs to other changes in the results page over that period rather than to AI Overviews alone. The keyword set is also defined by what triggered an AI Overview in December 2025, not in 2023.[1] Neither caveat changes the direction or the order of magnitude, and the earlier study, run over a single year with the same design, already found 34.5%.[2]
The click that never happens#
Clickthrough rate only describes the searches that still produce a click at all. That denominator is shrinking underneath it. SparkToro, working from Similarweb clickstream data, found that 68.01% of US Google searches ended without a click in the first four months of 2026, against 60.45% in 2024.[3] Put in absolute terms, only 276 of every 1,000 searches now reach the open web.
US Google searches that still send a visitor to the open web, first four months of 2026. SparkToro on Similarweb clickstream data.
The same dataset isolates the mechanism. Searches that trigger an AI Overview run at roughly 83% zero click, against about 60% for searches without one, and AI Overviews now appear on more than 20% of all searches.[3] A decade ago the zero click share was near 45%. The behaviour was already drifting before generative answers arrived. AI Overviews did not start the trend, they steepened it.
Two losses that compound#
These two findings are usually reported separately, which makes each one look survivable. They are not separate, and they multiply. Fewer searches produce a click at all, and the clicks that do survive are distributed to a page that captures a much smaller share of them. A brand can hold precisely the same position it held two years ago, publish nothing worse, lose no rankings, and still watch the traffic from that position fall by well over half without a single line in any report explaining why.
This is the shape of the problem that brings most companies to us. Nothing broke. No penalty landed. The dashboard shows stable rankings and falling sessions, and the usual diagnostic playbook has nothing to say about it, because the playbook assumes position is the thing being bought. It no longer is.
Why ranking harder does not fix it#
The instinct is to respond by competing harder for position, which is the one response the data rules out. In March 2026 Ahrefs examined 4 million URLs cited inside AI Overviews across 863,000 keyword SERPs and asked where those citations came from. Only 37.9% were drawn from Google's own top 10. A further 31.2% came from positions 11 to 100, and 31.0% came from pages that do not rank in the top 100 at all.[4]
Read that against the previous section and the trap becomes clear. Ranking in the top 10 is the only lever traditional SEO offers, it now returns 58% less traffic when an AI Overview appears, and it accounts for barely more than a third of the citations inside that AI Overview. Google is increasingly assembling its answers from the results of its own internal fan out queries rather than from the page you are optimising.[4]
A note on reconciling this with the other number in circulation. An August 2025 Ahrefs study of 15,000 prompts found that 80% of URLs cited by AI answers did not rank in Google's top 100 at all, which sounds like a contradiction.[6] It is not. That study measured AI answers across engines including ChatGPT and Perplexity, which have no reason to favour Google's index. The March 2026 study measures Google citing itself, which is the friendliest possible case for a ranking strategy. Even there, ranking buys you a minority of the citations. We wrote the longer version of how engines choose in how AI decides what to cite.
The demand moved, it did not vanish#
None of this means buyers stopped looking for what you sell. In March 2026, G2 surveyed 1,076 B2B software buyers and found 51% now begin product research with an AI chatbot more often than with Google, against 29% twelve months earlier.[5] In the same survey 69% said chatbot guidance led them to a different vendor than the one they set out to buy, and one in three bought from a vendor they had not heard of before the chatbot named it.[5]
That last figure is the one worth sitting with, because it cuts both ways. Being named by the engine can put you on a shortlist you were never going to reach. Being absent removes you from consideration by a buyer who will never learn you existed, and who leaves no trace in your analytics because they never visited. The full demand side case, including what this traffic converts at, is in the business case for GEO.
Budget has already followed. Conductor surveyed more than 250 leaders at US organisations with 500 or more employees and reports that 94% plan to increase answer engine optimisation investment in 2026, with an average of 12% of digital budget already allocated in 2025.[7] Conductor sells software in this category and the survey is US only, so read it as a directional signal from committed buyers rather than a market wide measurement. It still tells you what the teams you compete against are doing this year.
What actually moves a citation#
The foundational result here is academic rather than commercial. The Princeton, Georgia Tech and IIT Delhi team behind the original generative engine optimisation paper measured up to 40% gains in answer visibility across a 10,000 query benchmark, and the changes that produced those gains were unglamorous: adding concrete statistics, quotable sentences, and real citations to sources.[8] Writing that an engine can lift a defensible sentence out of is most of the work.
It is worth being equally clear about what does not carry the weight, because the category sells a lot of it. Structured data markup, llms.txt files, and entity declarations are hygiene at best. They are cheap, they occasionally help a crawler, and there is no controlled evidence that any of them moves citations. We went through the vendor claims one by one in can GEO be automated. If someone is selling you a schema audit as an AI visibility strategy, ask them for the before and after.
The harder half is measurement, and it is where most in house attempts stall. Engines answer the same question differently on repeated runs, so a single check tells you close to nothing and peer reviewed work on measuring AI visibility finds you need repeated runs per query, roughly seven or more, before a per brand estimate stabilises.[9] Multiply that by the number of engines your buyers actually use and the number of questions they actually ask, and it stops being something a person does on a Friday afternoon. That sampling problem is the reason our own numbers ship with confidence intervals and our methodology is public.
The strategic point underneath all of it: a click you lose to an AI Overview is gone for good, but a citation inside that answer is durable in a way a ranking never was. Consideration sets in AI answers are sticky, and claiming a slot while your category is still arguing about whether any of this is real is considerably cheaper than displacing an incumbent after it stops arguing.
See which questions in your category name a competitor instead of you, measured across engines rather than guessed.
Sources
- 58% lower clickthrough rate for the top ranking page when an AI Overview is present; position by position losses of 50.8%, 46.4%, 38.8%, 32.6% and 19.4%; 300,000 keywords, December 2023 against December 2025, published 4 February 2026: Ahrefs, “Update: AI Overviews Reduce Clicks by 58%”.
- The original measurement, 34.5% lower clickthrough rate at position 1 on informational keywords, 300,000 keywords, March 2024 against March 2025: Ahrefs, April 2025.
- 68.01% of US Google searches ended without a click in the first four months of 2026, against 60.45% in 2024; 276 of every 1,000 searches reach the open web; roughly 83% zero click on searches with an AI Overview against about 60% without: SparkToro on Similarweb clickstream data, reported by Search Engine Land, 2026.
- 37.9% of AI Overview citations come from Google's top 10, 31.2% from positions 11 to 100, 31.0% from outside the top 100; 4 million cited URLs across 863,000 keyword SERPs, data as of January 2026, published 2 March 2026: Ahrefs.
- 51% of B2B buyers start product research with an AI chatbot more often than Google, against 29% a year earlier; 69% switched vendor on chatbot guidance; one in three bought from a previously unknown vendor. Survey of 1,076 buyers fielded March 2026, published April 2026. G2 operates a software review marketplace: G2, “The Answer Economy”.
- 12% of AI cited URLs rank in Google's top 10 and 80% do not rank in the top 100, measured across AI answers from multiple engines on 15,000 prompts: Ahrefs, August 2025.
- 94% of respondents plan to increase answer engine optimisation investment in 2026; 12% average share of digital budget allocated in 2025. More than 250 C suite leaders, VPs and senior specialists at US organisations with 500 or more employees, across 12 or more industries. Vendor run survey, Conductor sells AEO software: Conductor, State of AEO and GEO, 2026.
- Content changes, specifically adding statistics, quotations and citations, lift visibility in generative answers by up to 40% on a 10,000 query benchmark: Aggarwal et al., “GEO: Generative Engine Optimization”, arXiv:2311.09735, KDD 2024.
- Repeated run sampling with confidence intervals as the minimum for a stable AI visibility estimate, roughly seven or more runs per query: Schulte, Bleeker and Kaufmann, “Don't Measure Once: Measuring Visibility in AI Search (GEO)”, arXiv:2604.07585, 2026.
Every figure above was traced to its primary source during a July 2026 research pass. Where a study is run by a vendor with a commercial interest in the result, it is labelled as such in the entry. The two Ahrefs clickthrough studies measure correlation, not causation, and their stated methodology limits are reproduced in the body rather than buried here. Noetio's own measurement methodology, including per engine sampling counts and confidence intervals, is published at noetio.com/methodology.