The best independent measurement of how AI summaries affect search behaviour comes from Pew Research Center, which tracked 68,879 real Google searches by 900 US adults in March 2025. Users clicked a traditional search result on 8 per cent of visits where an AI summary appeared, versus 15 per cent where one did not. Links inside the AI summary itself were clicked on about 1 per cent of visits. Browsing sessions ended after 26 per cent of pages with an AI summary, versus 16 per cent without, which suggests satisfied demand rather than only diverted demand.
Vendor studies corroborate the direction. Ahrefs, comparing 300,000 keywords between December 2023 and December 2025, found that click-through rate for the top-ranking result fell 78 per cent in raw terms where an AI Overview now appears (Ahrefs’ own adjusted figure, controlling for the parallel decline on non-AI keywords, is 58 per cent). Seer Interactive, analysing 3,119 informational queries across 42 client organisations, measured a 61 per cent decline in organic CTR and a 68 per cent decline in paid CTR. Both companies sell search software. The direction is consistent; the magnitudes are not directly comparable because the samples and methods differ.
The number that almost no coverage quotes is the one that matters most. On the same Ahrefs data, click-through rate for keywords that do not trigger an AI Overview also fell, from 7.6 per cent to 3.9 per cent, a decline of roughly 49 per cent. The AI Overview is an accelerant on a decline that is happening across the search channel as a whole.
Why the studies disagree, and what that tells you
Not every study finds the same thing. Semrush, tracking 200,000 keywords between January and October 2025, found that zero-click rates on AI Overview keywords actually fell slightly, from 33.75 per cent to 31.53 per cent. That contradicts the Ahrefs and Seer numbers on the surface, but the design difference explains it: Semrush compared the same keywords before and after an AI Overview appeared, while the cross-sectional studies compare keywords that have an AI Overview against keywords that do not. The second design confounds the AI Overview with query intent, because AI Overviews overwhelmingly appear on informational queries, which have always had poor click-through rates. Note the incentive inversion: Semrush also sells search software and is arguing against the alarming conclusion, which is a reason to take the finding seriously.
Google’s own position, stated by Liz Reid, VP and Head of Google Search, in August 2025, is that “total organic click volume from Google Search to websites has been relatively stable year-over-year” and that average click quality has increased. The post contains no percentages, no time series and no methodology, while criticising third parties for “flawed methodologies”.
The prediction that did not happen
Gartner predicted in February 2024 that traditional search engine volume would drop 25 per cent by 2026. It is now mid-2026. SparkToro’s measurement, using Datos clickstream data for US desktop, found that traditional search usage declined by less than 1 per cent over two and a half years. Searches per searcher actually grew. Meanwhile AI tool adoption rose from 8 per cent to 38 per cent of Americans, but the growth rate is slowing.
The honest reading is that AI use has been additive rather than substitutive so far, while the value of a search ranking has fallen sharply. Volume held; monetisable attention did not. Do not confuse a channel disappearing with a channel being repriced.
How AI systems choose what they cite
Only about 38 per cent of URLs cited in Google AI Overviews come from pages ranking in the top 10 for that query, down from roughly 76 per cent a year earlier. Semrush, in an earlier 2024 analysis, put the overlap lower still, at roughly 20 to 26 per cent. Two vendors disagree on magnitude but agree on direction: the majority of AI Overview citations do not come from the classic top 10.
Citation selection is probabilistic and unstable. A recent preprint decomposing 12,933 LLM responses across 20 brands, 8 languages and 3 models found that brand identity accounted for only 1.5 per cent of the variance in which brands an AI answer names. Resampling noise alone accounted for 34.8 per cent. The reliability of a single AI answer as a measurement is near 0.01, which means that most one-off AI visibility audits are measuring noise, not signal. This is a preprint and should be treated as suggestive rather than settled.
Cloudflare’s network-level data from June 2025 showed a crawl-to-refer ratio of approximately 70,900 to 1 for one major AI crawler: roughly 70,900 pages consumed for every referral visit sent back. Content is being read at enormous multiples of the traffic returned.
What the evidence for answer engine optimisation actually supports
The founding academic result on generative engine optimisation, published at ACM SIGKDD 2024, is real: adding verifiable statistics, credible quotations and source citations improved visibility in generative engine responses by up to 40 per cent on a synthetic position-adjusted metric. Two caveats that vendors routinely drop: the metric is a synthetic visibility score, not traffic or revenue, and the tests were run on 2023-2024 era generative engines.
A later academic benchmark, purpose-built to test conversational SEO methods across multiple domains, found that most such methods “are not only largely ineffective but also frequently have a negative impact on document ranking”. Traditional SEO strategies were significantly more effective. And gains diminish as more competitors adopt the same techniques, a zero-sum dynamic that means a technique which only works while few people use it is an arbitrage, not a system.
The durable part of the original finding, that content containing verifiable specifics gets selected more often, is simply a restatement of what makes content useful to a human reader. The mechanical tactics layered on top of it are the part that does not survive contact with competition.
How large the AI channel actually is
AI tools currently send less than 1 per cent of all outbound web traffic, a figure consistent with independent measurements from SparkToro, Ahrefs and DataReportal. That is the base rate that puts every “AI traffic up 400 per cent” headline in proportion.
Adobe Analytics data covering more than one trillion visits to US retail sites found that AI-sourced traffic rose 393 per cent year on year in Q1 2026 and converted 42 per cent better than non-AI traffic. Both numbers require context: Adobe does not publish AI traffic as a share of total traffic, so the growth rate has no interpretable denominator. And a visitor arriving from an AI assistant has typically already completed a research step, so they convert better for reasons that have nothing to do with the channel’s incremental value. Do not generalise a retail conversion advantage to a B2B advisory firm.
GWI’s survey of more than 240,000 people across 54 economies ranks AI tools fifth as a brand-discovery channel at 14.8 per cent, behind search engines at 32.4 per cent, TV advertising at 31.2 per cent and social advertising at 30.1 per cent. The channel is genuinely growing and currently small.
The concentration problem, and what to measure instead
Google holds 95.76 per cent of the search engine market in Africa. Bing holds 3.39 per cent. For a Mauritius-based or African firm, “search dependence” means dependence on one company, with no meaningful second engine to diversify into. Channel diversification in this market means leaving search, not switching engines.
The site-level damage is milder than page-level CTR numbers suggest. Across 19 member publishers surveyed by Digital Content Next, the median year-on-year decline in Google Search referral traffic was 10 per cent overall: 7 per cent for news brands and 14 per cent for non-news brands. That gap between a 58 per cent page-level CTR decline and a 10 per cent site-level traffic decline is itself informative: not every lost click translates into a lost visit.
The measurement that matters over the next 12 months is not page rankings but the mix: branded search share, direct and email arrival, and pipeline contribution. The Content Marketing Institute’s 2026 B2B survey captures the gap precisely: 87 per cent of B2B marketers using AI for content report improved productivity, but only 39 per cent report improved content performance. Producing more has not produced more effect. The firms that treat content as a system feeding multiple surfaces, rather than a volume game targeting a single ranking algorithm, are the ones whose systems will hold. The practical build sequence for that kind of system is covered in owned media systems that compound for B2B growth.
Key takeaways
- Users click a traditional search result on 8 per cent of visits where an AI summary appears, versus 15 per cent where one does not (Pew Research Center, March 2025, 68,879 searches).
- Click-through rates fell 49 per cent even on keywords that do not trigger an AI Overview, meaning the search channel as a whole is being repriced, not just the AI-affected portion.
- Gartner predicted a 25 per cent drop in search volume by 2026. Measured decline: under 1 per cent. AI use has been additive, not substitutive. Volume held; monetisable attention did not.
- Most conversational SEO methods are “largely ineffective” or actively harmful in academic testing, and gains diminish as competitors adopt the same techniques.
- AI tools send less than 1 per cent of all outbound web traffic. The channel is growing fast off a small base.
- Google holds 95.76 per cent of search in Africa. In this market, channel diversification means leaving search, not switching engines.
Frequently asked questions
How much does an AI Overview reduce clicks on search results?
Pew Research Center found users clicked a traditional result on 8 per cent of visits where an AI summary appeared, versus 15 per cent where one did not. Vendor studies report declines of 34 to 68 per cent in click-through rate for the top result, though the studies use different samples, methods and time periods, which is why the figures do not converge.
Is search volume actually declining because of AI?
No. Gartner predicted a 25 per cent decline by 2026, but SparkToro’s clickstream measurement found traditional search usage fell less than 1 per cent over two and a half years, and searches per searcher grew. AI adoption rose from 8 to 38 per cent of Americans but has been additive rather than substitutive. The value of a ranking is falling; the volume of searches is not.
Does optimising content for AI answer engines work?
The evidence is mixed. The original academic study (GEO, ACM SIGKDD 2024) found that adding verifiable statistics and citations improved a synthetic visibility score by up to 40 per cent. A later benchmark found most conversational SEO methods ineffective or actively harmful, and that gains diminish as competitors adopt them. The durable part is that content with verifiable specifics gets selected more often, which is simply what makes content useful to any reader.
How much web traffic do AI tools actually send?
Less than 1 per cent of all outbound web traffic, consistent with independent measurements from SparkToro, Ahrefs and DataReportal. Adobe found AI-sourced traffic to US retailers rose 393 per cent year on year in Q1 2026, but the growth rate is off an undisclosed base. GWI ranks AI tools fifth as a brand-discovery channel at 14.8 per cent, behind search, TV and social advertising.
What should firms measure instead of page rankings?
Branded search share, direct and email arrival rates, and pipeline contribution over a 12-month window. Site-level traffic decline (median 10 per cent across 19 large publishers) is considerably milder than page-level CTR decline (58 to 68 per cent), which means not every lost click translates into a lost visit. The measurement shift is from ranking position to audience relationship strength.
Sources
- Pew Research Center, “Google users are less likely to click on links when an AI summary appears in the results”, 22 July 2025. pewresearch.org
- Ahrefs, “Update: AI Overviews Reduce Clicks by 58%”, 4 February 2026. ahrefs.com
- Seer Interactive, “AIO Impact on Google CTR: September 2025 Update”. seerinteractive.com
- Semrush, “Semrush AI Overviews Study”, refreshed 15 December 2025. semrush.com
- Liz Reid, “AI in Search is driving more queries and higher quality clicks”, Google blog, 6 August 2025. blog.google
- Gartner, “Gartner Predicts Search Engine Volume Will Drop 25% by 2026”, 19 February 2024. gartner.com
- SparkToro, “New Research: 20% of Americans use AI tools 10X+/month”, 26 August 2025. sparktoro.com
- SparkToro, “In 2026, Less than One Third of Google Searches Still Send a Click”, 9 June 2026. sparktoro.com
- Ahrefs, “Update: 38% of AI Overview Citations Pull From The Top 10”, 2 March 2026. ahrefs.com
- Dmitrij Zatuchin, “Where Does the Noise Come From? A Variance-Components Decomposition of Non-Determinism in LLM Brand Answers”, arXiv 2607.13304, 14 July 2026. arxiv.org
- Cloudflare, “The crawl before the fall: understanding AI’s impact on content providers”, 1 July 2025. blog.cloudflare.com
- Aggarwal et al., “GEO: Generative Engine Optimization”, ACM SIGKDD 2024. arxiv.org
- Puerto et al., “C-SEO Bench: Does Conversational SEO Work?”, arXiv 2506.11097, revised October 2025. arxiv.org
- Adobe Analytics, AI traffic trends report Q1 2026, reported by TechCrunch. techcrunch.com
- GWI / DataReportal, “Digital 2026 Mid-Year Global Update Report”, April 2026. datareportal.com
- StatCounter Global Stats, Search Engine Market Share Africa, June 2026. gs.statcounter.com
- Digital Content Next, reported via Digiday. digiday.com
- Content Marketing Institute and MarketingProfs, “B2B Content and Marketing Trends: Insights for 2026”, survey of 1,015 B2B marketers. contentmarketinginstitute.com
