The Macro Shift
The Great Unbundling
Since Google's AI Overviews launched in May 2024, a rising share of search sessions now end inside a synthesized answer instead of a click to your website, and the strongest data available says that shift is real, accelerating, and still only partly measured.
Part of The Macro Shift in the Insights library.
Abstract
Search is being unbundled from the click. Google's own numbers show its answer layer scaling to billions of monthly users in under two years, Pew's panel data shows people clicking through at roughly half the old rate when an AI summary appears, and independent clickstream trackers put the US zero-click rate at 68% in early 2026, up from about 60% in 2024. None of that is seriously in dispute. What is contested is the size of the effect, whether it is a new phenomenon or an old trend picking up speed, and how much of the migrated attention has anywhere solid to land, because no independent census of the answer layer itself exists yet.
The search results page is turning into a page you sometimes never see. Google reported its AI Overviews reached 2 billion monthly users across more than 200 countries and territories by mid-2025, and AI Mode, which launched to 100 million monthly users in the US and India in July 2025, crossed 1 billion monthly users by Google I/O 2026, with query volume more than doubling every quarter along the way. Google frames this as additive, telling investors AI Overviews drive more than 10% more query volume on the query types where they appear, and that total Search query volume is at an all-time high. Outside Google, the same shift shows up as fast growth in standalone conversational tools: ChatGPT's self-reported weekly active users climbed from 100 million in November 2023 to 900 million by February 2026, and DataReportal's global estimate, once China is included, puts more than a billion people on standalone AI platforms monthly. At the same time, the ground underneath all of this, plain search-engine use, is quietly eroding: GWI survey data cited by DataReportal found only 80.3% of online adults worldwide used a search engine in the past month, the lowest figure on record for that question, down about 2.5% from the year before. Search hasn't been replaced. It has been joined by a second, faster-growing layer that increasingly gets to the answer first.
What that does to buyer behavior is the part with the sharpest, most granular data behind it. Pew Research tracked the actual browsing of 900 US adults across 68,879 real Google searches in March 2025, and found people clicked through to a traditional result link in only 8% of searches where an AI summary appeared, versus 15% where it didn't, roughly half the rate. Clicking a link inside the summary itself happened in just 1% of visits, and 26% of AI-summary searches ended with no further browsing at all, compared with 16% for ordinary searches. Ahrefs' separate rank-tracking data, comparing December 2023 to December 2025, found a 58% lower average click-through rate on the very top organic result when an AI summary is present. People are still asking questions, often more of them, but a growing share are getting satisfied before they ever reach a business's own page. And the audience doing this first is skewing young: the Reuters Institute found 7% of people use AI chatbots weekly for news, rising to 15% among under-25s, with ChatGPT the single most-used tool. A National Bureau of Economic Research analysis of actual ChatGPT conversations found that practical guidance, information-seeking, and writing together make up roughly 80% of what people use it for, and that the share of usage that is personal rather than work-related has been climbing for years. The buying journey isn't disappearing: it's compressing, and a meaningful part of it now happens inside a chat window before a brand's own website ever enters the picture.
This is exactly the terrain a Visibility Corpus exists to read. If attention is migrating from ranked links to synthesized answers, then knowing where your specific market's queries sit today, and how fast they're moving, matters more than any single traffic number in this study, because none of the figures here are about your category or your city. Google's own referral share is still the largest single source of traffic by far: Ahrefs' live tracker puts it at roughly a quarter of tracked referrals, even amid a declining trend. Retail is already showing a different pattern. Adobe measured generative-AI-referred traffic to US retail sites up 1,300% year over year across the November-December 2024 holiday period, and roughly doubling every two months since. That is a small base growing extremely fast, which is exactly the signature of a wedge you want to catch early rather than after it's obvious. The corpus is the map of where your buyers' attention actually sits right now, across the open web and the answer layer both; Search Surface Optimization is the discipline of winning presence on that terrain as it's mapped, not as it looked two years ago; and a Machine-Readiness Score is how you turn that from a hunch into a number you can track quarter over quarter. The unbundling described in this study is the macro current. Where it is pulling your specific market is a separate, answerable question.
The data, in one read
The Answer Layer Went From Feature to Default
Two years ago, an AI-written summary at the top of a Google results page was a limited experiment. It is now one of the largest consumer software rollouts on record. Google told investors in its Q2 2025 earnings call that AI Overviews had reached 2 billion monthly users across more than 200 countries and territories, up from 1.5 billion just two months earlier in May 2025. AI Mode, Google's more conversational, multi-turn search experience, launched to 100 million monthly users across the US and India in July 2025 and crossed 1 billion monthly users by Google I/O 2026, roughly a year after debut, with query volume more than doubling every quarter since launch.
Google's framing of this growth is that it is additive, not cannibalizing. The company points to Sundar Pichai's remark that AI Overviews drive more than 10% additional query volume on the query types where they appear, and it says total Search query volume is at an all-time high. Both things can be true at once: more total searches happening, and a larger share of them resolving without a visit to any external website. The rollout speed itself is the least contested fact in this study: it comes directly from Google's own disclosures. What people do once the answer appears is a separate, and far more argued-over, question.
AI Mode went from 100 million monthly users to more than a billion in about a year, with query volume more than doubling every quarter since launch.
What Happens to the Click When an Answer Appears
The most granular, credible behavioral data available comes from Pew Research Center, which in March 2025 tracked the actual browsing activity of a panel of 900 US adults across 68,879 real Google searches, 12,593 of which produced an AI summary. The results are stark: people clicked through to a traditional search result link in only 8% of AI-summary searches, versus 15% of searches without one, roughly half the rate. Clicking a link placed inside the AI summary itself happened in just 1% of visits. And 26% of AI-summary searches ended in outright abandonment, no further browsing at all, compared with 16% for searches without a summary.
This is not a single anecdote. Ahrefs' independent rank-tracking data, comparing December 2023 (before AI Overviews existed) to December 2025, found a 58% lower average click-through rate on the number-one organic result specifically when an AI summary is present. Two different methodologies, a real-user panel and a rank-tracking dataset, point the same direction: being cited inside an AI summary is not the same as being visited, and for many searches now, the summary is where the interaction ends.
People clicked a link inside the AI summary itself in just 1% of visits, and more than a quarter of AI-summary searches ended with no further browsing at all.
Google Disputes the Pew Findings
Google did not let the Pew study stand unanswered. In a statement reported by Search Engine Land, Google said the research "uses a flawed methodology and skewed queryset that is not representative of Search traffic," reiterated that it sends billions of clicks to websites every day, and said it has not observed significant drops in aggregate web traffic. Google pointed back to Pichai's 10%-plus query-volume figure as its counter-evidence.
This is worth taking seriously as a live, unresolved dispute rather than settling it in either direction. Google did not publish its own click-through data to substantiate the rebuttal, and Pew's panel methodology, while limited to US English-language searches on one platform, is transparent and independently replicable in a way a company's internal aggregate figures are not. The read is that both positions can be partially correct: Google may genuinely be sending more total clicks in absolute terms while the per-search probability of a click keeps falling, the same way a store can sell more units at a lower margin. Until a neutral third party reconciles the two, treat this specific disagreement as exactly what it is: a contested claim, not a resolved one.
The Zero-Click Line Is Older Than AI, and It Is Steepening
The rise of the AI summary did not create zero-click search. It accelerated a trend that has been building for a decade. SparkToro's synthesis of clickstream panel data (Similarweb and Datos/Semrush) traces the US Google zero-click rate, the share of searches ending without any click at all, from roughly 45% in 2016, to 49% in 2019, to 60.45% in 2024, to 68.01% across the first four months of 2026. Within just that 2024-to-2026 window, the share of searches producing any click fell 9.51 percentage points, a 22.9% relative decline, while repeat-search behavior (searching again without clicking) rose 7.2 points.
One detail in that same dataset complicates any simple story: Google's own AI Mode accounted for only 0.34% of all searches tracked in the first four months of 2026, despite having crossed a billion monthly users. The zero-click increase is not simply people switching from classic search results to AI Mode results. It is a broader shift in how the results page itself behaves: featured snippets, knowledge panels, Maps and YouTube redirection, and now AI Overviews, all compounding on top of a rising paid-click share that predates any of this. No dataset located for this study cleanly isolates how much of the recent acceleration is attributable to AI Overviews specifically versus this decade-long structural trend continuing on its own trajectory. That attribution gap is, not coincidentally, exactly where Google's rebuttal to Pew lives.
Where the Traffic Actually Lands
If fewer searches produce a click, where does the traffic that still moves actually go? Ahrefs' live, opt-in panel tracker (chatgpt-vs-google.com) shows Google's share of referral traffic trending downward across 2025 and 2026, but the decline is volatile rather than smooth, including a 7.7% jump in February 2026 within an overall downward drift. At last check, Google still held roughly a quarter of all tracked referral traffic, by a wide margin the single largest source measured.
Standalone AI tools, meanwhile, are not picking up the slack in raw traffic terms. The same Ahrefs tracker finds that ChatGPT, Perplexity, and similar tools collectively send less than 1% of all web traffic out to other sites, a strikingly small number given that these platforms now count their users in the hundreds of millions to over a billion. People are spending time inside these tools, but for most use cases they are not clicking out of them the way search engines historically sent people onward. That gap between attention captured and traffic released is one of the more important, underdiscussed facts in this entire study: the audience is there, but the referral pipe most businesses built their marketing around is not.
Retail Is the Early Exception
One category looks meaningfully different from the general pattern: shopping. Adobe Analytics measured traffic to US retail websites arriving from generative AI sources up 1,300% year over year across the November-December 2024 holiday period, spiking to 1,950% year over year on Cyber Monday 2024 specifically, and up 1,200% in February 2025 compared with a July 2024 baseline, a growth curve Adobe characterized as roughly doubling every two months since September 2024.
Adobe's reporting on the 2025 holiday season describes continued triple-digit-plus growth in the same direction, though the specific percentages in that later release could not be independently re-verified for this study and are treated here as emerging rather than established. The pattern to take from this is directional, not precise: growth off a genuinely small base can produce extraordinary percentages without yet representing a majority channel, but a trend compounding this fast, in a category as commercially important as retail, is worth tracking closely well before it becomes obvious in your own analytics.
Who Is Actually Leaving Search, and For What
The demographic and use-case data available paints a more specific picture than "AI is replacing search" in general. The Reuters Institute's 2025 Digital News Report found that 7% of people use AI chatbots weekly for news overall, but that figure rises to 15% among people under 25, with ChatGPT the most-used tool at 4% weekly, ahead of any other AI source, including Google's own AI Overviews as a news destination, a comparatively low showing the report calls notable given how prominently AI Overviews sit atop search results.
A privacy-preserving analysis of actual ChatGPT conversations by researchers publishing through the National Bureau of Economic Research found that practical guidance, information-seeking, and writing together account for roughly 80% of all conversations on the platform, and documented a rising non-work share of usage, from around half of all messages toward more than 70% across the paper's coverage window from November 2022 to July 2025 (two of the paper's seven authors are OpenAI-affiliated, a disclosed limitation on independence worth noting). Meanwhile, at the population level, DataReportal's citation of GWI survey data found that only 80.3% of online adults worldwide reported using a search engine in the past month, the lowest figure GWI has recorded for that question, a relative decline of about 2.5% over the prior year. Search remains, by a wide margin, the top global channel for brand discovery and purchase research, and global search ad spend was projected at $352 billion for 2025, up 11.1% year over year. The shift is real and it is concentrated among younger users and certain use cases, but it has not dethroned search as the default starting point for most of the world.
A Prediction That Hasn't Cleanly Landed
In February 2024, Gartner formally predicted that traditional search engine volume would fall 25% by 2026 as generative AI chatbots and virtual agents became substitute answer engines. Two years later, the directional evidence in this study points the same way Gartner predicted: the zero-click rise, the referral-share decline, the standalone-AI adoption curve. But no dataset located for this study isolates a clean, attributable 25% contraction in search-engine volume specifically, and search-industry press has publicly challenged the prediction's methodology as its own deadline arrived.
The way to hold this: Gartner called the direction early and got the shape of the shift largely right, but its precise magnitude still can't be verified against the data available today. That is a useful pattern to recognize across this whole body of evidence: forecasts and vendor trackers routinely get the direction of change right well before anyone can put a defensible number on its size.
Winning Inside the Answer Is a Different Discipline
If a growing share of buyer attention resolves inside a synthesized answer rather than a clicked page, the practical question becomes: can you influence what that answer says about you? The peer-reviewed paper that coined the term Generative Engine Optimization, published by researchers at Princeton, the Allen Institute for AI, Georgia Tech, and IIT Delhi at ACM SIGKDD 2024, built a large benchmark across nine domains and found that specific content strategies, adding citations and statistics among them, could lift a source's visibility inside generative-engine responses by up to 40% in controlled benchmark experiments.
That figure comes from a controlled academic benchmark, not a live re-run against today's production systems, so it should be read as evidence that the answer layer is influenceable in principle, not a guaranteed lift for any specific brand. Still, it is the clearest existing signal available that writing for the surfaces engines actually pull from is a distinct discipline from writing for a ranking algorithm: structured facts, citable statistics, clear sourcing, not just the right keywords. Most businesses have not yet built a practice around it.
The evidence, in numbers
Key findings, dated and sourced
-
Google's AI Overviews reached 2 billion monthly users across 200+ countries and territories, up from 1.5 billion in May 2025; AI Overviews drive over 10% more query volume globally on the query types where they appear.
established Google (Alphabet), via Q2 2025 earnings call, TechCrunch: "Google's AI Overviews have 2B monthly users, AI Mode 100M in the US and India" (2025-07-23)
-
Google's AI Mode grew from 100 million monthly active users (US and India, July 2025) to more than 1 billion monthly users by Google I/O 2026, about a year after debut, with query volume more than doubling every quarter, pushing total Search query volume to an all-time high.
established Google I/O 2026 Search keynote recap (2026-05)
-
In a panel of 900 US adults tracked in March 2025 (68,879 Google searches, 12,593 producing an AI summary), users clicked a traditional result link in 8% of AI-summary searches versus 15% of non-summary searches, clicked a link inside the AI summary itself in only 1% of visits, and abandoned the session (no further browsing) in 26% of AI-summary visits versus 16% without.
established Pew Research Center (Athena Chapekis and Anna Lieb), "Do people click on links in Google AI summaries?" (2025-07-22)
-
Google publicly disputed the Pew Research findings, stating the study "uses a flawed methodology and skewed queryset that is not representative of Search traffic," that it directs billions of clicks to websites daily, and that it has not observed significant drops in aggregate web traffic.
contested Google, statement reported by Search Engine Land, "Google's AI Overviews are hurting clicks: Pew study" (2025-07-22)
-
US Google zero-click search rate (searches ending without any click) rose from roughly 45% in 2016 and 49% in 2019, to 60.45% in 2024, to 68.01% across the first four months of 2026.
emerging SparkToro (Rand Fishkin), citing Similarweb and Datos/Semrush panel data, "In 2026, Less than One Third of Google Searches Still Send a Click" (2026-06)
-
Within the 2024-to-2026 US clickstream comparison, the share of searches producing any click fell 9.51 percentage points (a 22.9% relative decline) while repeat-search behavior rose 7.2 points; Google's own AI Mode accounted for only 0.34% of searches in Jan-Apr 2026 despite its billion-plus user base; separately, standalone AI tools (ChatGPT, Perplexity, etc.) send less than 1% of all web traffic out.
emerging SparkToro / Similarweb / Ahrefs, "In 2026, Less than One Third of Google Searches Still Send a Click" (2026-06)
-
Google's share of referral traffic has trended down over the 2025-2026 period, though month-to-month movement is volatile rather than a smooth decline (e.g., a 7.7% jump in February 2026 amid an overall downward drift); Google still holds roughly a quarter of tracked referral traffic, the largest single source.
emerging Ahrefs ChatGPT-vs-Google traffic tracker (live, point-in-time) (2026-06)
-
The presence of an AI summary correlates with a 58% lower average click-through rate for top-ranking (position-1) organic pages, comparing December 2023 (pre-AI Overviews) to December 2025.
emerging Ahrefs AI Overviews CTR-reduction update (2025-12)
-
Gartner formally predicted in February 2024 that traditional search engine volume would fall 25% by 2026 as generative AI chatbots and virtual agents became substitute answer engines. By mid-2026, directional evidence confirms a real and accelerating shift, but no dataset isolates a clean, attributable 25% search-volume contraction, and search-industry press has publicly challenged the prediction's methodology as the deadline arrived.
contested Gartner, Inc. (Alan Antin, VP Analyst), "Gartner Predicts Search Engine Volume Will Drop 25% by 2026, Due to AI Chatbots and Other Virtual Agents" (2024-02-19)
-
Traffic to US retail websites from generative AI sources rose 1,300% year over year across the Nov-Dec 2024 holiday period, with a 1,950% year-over-year spike on Cyber Monday 2024, and 1,200% in February 2025 versus a July 2024 baseline (a 7-month comparison), a pattern Adobe describes as roughly doubling every two months since September 2024.
established Adobe (Adobe Analytics), "Adobe Analytics: Traffic to U.S. retail websites from Generative AI sources jumps 1,200 percent" (2025-03-17)
-
Adobe reported continued growth in generative-AI-referred retail traffic and rising consumer adoption of AI-assisted shopping research for the 2025 US holiday season, consistent in direction with its earlier 2025 reporting, though the specific magnitudes could not be independently re-confirmed for this study.
emerging Adobe (Adobe Analytics), "Generative AI-Powered Shopping Rises with Traffic to U.S. Retail Sites" (2025-12)
-
7% of people use AI chatbots weekly for news overall, rising to 15% among under-25s, with ChatGPT the most-used tool (4% weekly), ahead of other AI news sources; Google's AI Overviews show a comparatively low showing as a news source given their prominence atop search results.
established Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025, Executive Summary (2025-06)
-
A privacy-preserving analysis of ChatGPT conversations found Practical Guidance, Seeking Information, and Writing together account for roughly 80% of all conversations, and documented a rising non-work share of usage across the paper's Nov 2022-Jul 2025 coverage window, from roughly half of messages toward more than 70%.
emerging National Bureau of Economic Research (Chatterji, Cunningham, Deming, Hitzig, Ong, Shan, Wadman), NBER Working Paper No. 34255, "How People Use ChatGPT" (2025-09)
-
Only 80.3% of online adults worldwide report using a search engine within the past month, the lowest figure on record for this question, a relative decline of about 2.5% (-210 basis points) over the prior 12 months; search remains the top global brand-discovery channel, and global search ad spend was projected at $352 billion for 2025, up 11.1% year over year.
established DataReportal / Kepios (Simon Kemp), based on GWI survey data, Digital 2026: Global Overview Report (2025-10)
-
More than 1 billion people worldwide now use standalone AI platforms monthly once China is included, a composite estimate drawing on divergent methodologies (third-party panel MAU, OpenAI's self-reported weekly actives, and a separate China adoption estimate); no single authoritative AI-usage census yet exists.
contested DataReportal / Kepios, citing Similarweb and a China AI-adoption report, Digital 2026: Global Overview Report (2025-10)
-
ChatGPT's self-reported weekly active user count climbed from 100 million (Nov 2023) to 400 million (Feb 2025) to 700 million (Jul 2025) to 800 million (Oct 2025) to 900 million (Feb 2026), with January-February 2026 on track to be OpenAI's largest-ever months for new subscribers.
established OpenAI (Sam Altman), reported via TechCrunch, "ChatGPT reaches 900M weekly active users" (2026-02-27)
-
The peer-reviewed paper that coined the term "Generative Engine Optimization" built a large-scale, nine-domain benchmark (GEO-BENCH) and demonstrated that specific content strategies, such as adding citations and statistics, can boost a source's visibility in generative-engine responses by up to 40% in controlled benchmark experiments.
established Princeton University / Allen Institute for AI / Georgia Tech / IIT Delhi (Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande); published ACM KDD 2024, "GEO: Generative Engine Optimization," arXiv:2311.09735 (ACM SIGKDD 2024) (2023-11-16)
Learning outcomes
What this study teaches
- A synthesized answer is now a normal outcome of a search, not an edge case, for a large share of query types. Your website's job is no longer only to rank, it also has to be the kind of source an answer draws from and credits.
- Being cited inside an AI summary is not the same as being visited. Pew found people click a link inside the summary itself in only 1% of cases, so visibility inside the answer layer and traffic to your site are two different things to track separately.
- The click decline predates generative AI by close to a decade and is accelerating on top of that trend, not starting from zero. Treat any single dip in referral traffic as one data point in a longer pattern, not proof of one cause you can fix with one tactic.
- If you sell online, watch the shopping category closely. Generative-AI-referred retail traffic is growing off a small base but at extraordinary rates, the kind of early signal worth acting on before it shows up clearly in your own analytics.
- Every measurement source in this space, including Google itself, disagrees with at least one other on magnitude. Treat any headline AI-search statistic as directional until you have your own read on where your specific market's attention is actually moving.
Honest limits
What this does not yet settle
- No independent, standardized census of the AI-answer layer exists. There is no equivalent of a GWI or Pew survey, or a public, audited share-of-answer metric across generative engines. Every AI-summary CTR and prevalence figure beyond Pew's panel study comes from a proprietary vendor tracker (Ahrefs, Similarweb) whose full panel and methodology are not independently published, so exact percentages cannot currently be cross-checked against each other.
- Attribution between AI Overviews and the pre-existing zero-click trend is unresolved. No dataset here isolates how much of the rising zero-click rate is caused specifically by AI Overviews and AI Mode versus a decade-long trend already driven by featured snippets, knowledge panels, and Maps or YouTube redirection. This is precisely the gap Google's rebuttal to Pew rests on, and no third party has published a controlled decomposition.
- Geographic coverage outside the US, UK, and a handful of EU markets is thin. SparkToro's planned breakdowns for the UK, Italy, Germany, France, and Canada had not been published at the time of this research, and no source located here gives comparably rigorous figures for India, Southeast Asia, Latin America, or the Middle East, despite Google stating AI Overviews cover 40-plus languages including Arabic, Chinese, Malay, and Urdu.
- Publisher-level revenue impact, as opposed to traffic, is documented only anecdotally rather than through an industry-wide, audited study. SparkToro itself notes that traffic can fall sharply even as revenue rises for some site owners, so none of the traffic figures in this study should be read as a direct stand-in for revenue impact.
- AI-platform user-count methodologies diverge sharply: OpenAI's self-reported weekly actives, Similarweb and Kepios's third-party monthly-active panel estimates, and separate China adoption figures do not share a common measurement standard. Any single global 'AI user' total should be read as a range estimate, not a precise figure, until a neutral party reconciles the methods.
- No large study located here directly measures whether AI-chatbot query volume represents net-new demand or substitution of search-engine query volume. The NBER paper describes what people ask ChatGPT but does not establish a causal displacement estimate against Google search.
This is a synthesis of dated, attributed evidence, not a census. The AI-answer layer in particular has no independent, Nielsen-grade measurement yet, so readings of it are directional and named as a frontier, never presented as settled.
Straight answers
Frequently asked questions
Are people really clicking search results less often when an AI summary shows up?
Yes. Pew Research Center tracked a panel of 900 US adults across 68,879 real Google searches in March 2025 and found people clicked through to a traditional result link in only 8% of searches where an AI summary appeared, versus 15% where it didn't, roughly half the rate. Clicking a link inside the summary itself happened in just 1% of visits, and 26% of AI-summary searches ended with no further browsing at all. Ahrefs' separate rank-tracking data found a similar pattern: a 58% lower average click-through rate on the top organic result when an AI summary is present, comparing December 2023 to December 2025.
How high has zero-click search actually gotten?
SparkToro's synthesis of clickstream panel data puts the US Google zero-click rate, the share of searches ending without any click, at 68.01% across the first four months of 2026, up from 60.45% in 2024. That rate has been climbing for a decade, from roughly 45% in 2016 to 49% in 2019, so the recent jump is an acceleration of an older trend rather than something that started with AI Overviews. This is emerging-tier vendor-panel data, not an audited census, so treat the exact percentages as directional.
Does Google agree that AI Overviews are cutting into website clicks?
No, and this is a genuinely unresolved dispute worth naming plainly. Google publicly rebutted the Pew study, saying it "uses a flawed methodology and skewed queryset that is not representative of Search traffic," and pointed to its own claim that AI Overviews drive more than 10% additional query volume on the query types where they appear, without publishing its own click-through data to back up the rebuttal. The read in the study is that both sides can be partly right at once: Google may be sending more total clicks while the odds of any single search producing a click keep falling.
Is the shift to AI answers hurting all businesses equally, or is one category different?
Retail stands out as an early exception. Adobe Analytics measured generative-AI-referred traffic to US retail websites up 1,300% year over year across the November-December 2024 holiday period, spiking to 1,950% on Cyber Monday 2024, and roughly doubling every two months since. That growth is off a small base, so it doesn't yet represent a majority traffic channel, but the study flags it as the kind of early signal worth watching before it shows up clearly in your own analytics.
Can a business actually influence what an AI answer says about it, or is that still unproven?
There is real evidence it's influenceable, but it comes with a caveat worth stating plainly. The peer-reviewed paper that coined the term Generative Engine Optimization, built by researchers at Princeton, the Allen Institute for AI, Georgia Tech, and IIT Delhi, found that specific content strategies, like adding citations and statistics, could lift a source's visibility inside generative-engine responses by up to 40% in a controlled academic benchmark. That figure is from a controlled benchmark, not a live test against today's production AI systems, so it shows the answer layer can be influenced in principle rather than guaranteeing a lift for any specific brand.
Provenance
References
- TechCrunch, "Google's AI Overviews have 2B monthly users, AI Mode 100M in the US and India" (2025) https://techcrunch.com/2025/07/23/googles-ai-overviews-have-2b-monthly-users-ai-mode-100m-in-the-us-and-india
- Google, I/O 2026 Search keynote recap https://blog.google/products-and-platforms/products/search/search-io-2026/
- Pew Research Center, "Do people click on links in Google AI summaries?" (2025) https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
- Search Engine Land, "Google's AI Overviews are hurting clicks: Pew study" (2025) https://searchengineland.com/google-ai-overviews-hurting-clicks-study-459434
- SparkToro, "In 2026, Less than One Third of Google Searches Still Send a Click" https://sparktoro.com/blog/in-2026-less-than-one-third-of-google-searches-still-send-a-click/
- Ahrefs, ChatGPT-vs-Google traffic tracker (live) https://chatgpt-vs-google.com/
- Ahrefs, AI Overviews CTR-reduction update https://ahrefs.com/blog/ai-overviews-reduce-clicks-update/
- Gartner, Inc., "Gartner Predicts Search Engine Volume Will Drop 25% by 2026" (2024) https://www.gartner.com/en/newsroom/press-releases/2024-02-19-gartner-predicts-search-engine-volume-will-drop-25-percent-by-2026-due-to-ai-chatbots-and-other-virtual-agents
- Adobe Analytics, "Traffic to U.S. retail websites from Generative AI sources jumps 1,200 percent" (2025) https://blog.adobe.com/en/publish/2025/03/17/adobe-analytics-traffic-to-us-retail-websites-from-generative-ai-sources-jumps-1200-percent
- Adobe, "Generative AI-Powered Shopping Rises with Traffic to U.S. Retail Sites" (2025) https://business.adobe.com/blog/generative-ai-powered-shopping-rises-with-traffic-to-retail-sites
- Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025 https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2025/dnr-executive-summary
- National Bureau of Economic Research, Working Paper No. 34255, "How People Use ChatGPT" https://www.nber.org/papers/w34255
- DataReportal / Kepios, Digital 2026: Global Overview Report https://datareportal.com/reports/digital-2026-global-overview-report
- TechCrunch, "ChatGPT reaches 900M weekly active users" (2026) https://techcrunch.com/2026/02/27/chatgpt-reaches-900m-weekly-active-users
- Aggarwal et al., "GEO: Generative Engine Optimization," arXiv:2311.09735 (ACM SIGKDD 2024) https://arxiv.org/abs/2311.09735
Every measured figure is dated to its capture and tagged with an evidence tier. Every cited work is real and locatable. Where an engine could not be captured this round, it is named as uncaptured, not estimated. Small-sample readings are labelled as directional.