Global & MSME Commerce
The Digital Divide 2.0
Buyer attention is verifiably moving from links to synthesized answers, but whether that shift is quietly widening the gap between small and large businesses is a question no public study has yet answered, which is exactly why it matters.
Part of MSME & Global Commerce in the Insights library.
Abstract
When Google shows an AI Overview, searchers click through to a traditional result in only 8 percent of visits, versus 15 percent when no summary appears, according to Pew Research Center's study of real Google search sessions. That confirms the first half of this study's question: attention is measurably migrating from clickable links to synthesized answers. The second half, whether small businesses specifically fall behind larger competitors inside those answers because their sites are harder for a machine to read, has never been measured by any public source we could locate, so we are naming it as an open question rather than reporting it as settled fact.
The surface is changing on two timelines at once, and only one of them is solidly measured. The rollout is a matter of record: Google launched AI Overviews in the US in May 2024, had them running in more than 100 countries by October 2024, and reached over 200 countries and territories in 40-plus languages by May 2025, everywhere except Cuba, Iran, and mainland China. In 2026, Google went further and put its position in writing, stating in its own documentation that optimizing for generative AI search is, in its words, still SEO. That is the first time the company has said this formally rather than at a conference. What is less settled is exactly how much of today's search volume the feature touches. A Wikipedia compilation citing an unnamed third party tracker puts AI Overview prevalence at roughly 48 percent of Google queries in March 2026, up from about 6.49 percent a year earlier. That is a striking trajectory, but because the underlying tracking method and vendor are not disclosed, we treat the specific figure as directional rather than confirmed, even while the direction of travel is not in doubt.
What buyers do once an answer appears is better documented than how often it appears. Pew's fielded data on actual Google searches, 900 US adults and 68,879 unique queries collected in March 2025, found the click-through rate to a traditional link drops from 15 percent to 8 percent the moment an AI summary shows up on the page. Separately, a Botify and Demandsphere study found that when an AI Overview and a featured snippet appear together, they can occupy roughly 67.1 percent of visible desktop screen space and 75.7 percent on mobile, meaning even the top organic result can sit below the fold; we flag this as a single study rather than an industry consensus. Trust is following a similar downward path for the answers those engines produce: the Reuters Institute's 48-market Digital News Report found audiences expect AI to make news less trustworthy, by a net score of negative 18, and less accurate, by negative 8, even as weekly use of AI chatbots for news has reached 7 percent globally and 15 percent among people under 25. Buyers are clicking less and trusting the summaries less, at the same time, which is the paradox any small business now has to plan around.
If fewer clicks are reaching individual websites, the natural next question is where the attention that used to land there is actually going, and no public source has published a firm-size breakdown of that yet. What we do have is a hint about which sources AI answers lean on: a Semrush analysis of 200,000 US keywords that triggered a Google AI Overview found Quora was the most frequently cited source, with Reddit second, meaning large user-generated content platforms, not individual business websites, dominated that sample's citations regardless of any one business's own site quality. That single finding is the clearest signal in this entire evidence set for why a map of where attention actually sits, not just where your own site ranks, is the more useful unit of analysis going forward. It is also exactly the kind of question a Visibility Corpus is built to answer: not whether your page is optimized in isolation, but where your market's attention is distributed across search, social, review platforms, and AI answers, and how that distribution is moving.
The data, in one read
The Click Is Already Leaving the Building
Start with the one number in this study that is about as close to settled fact as web behavior research gets. Pew Research Center did not survey people about their opinions of AI search; it watched what 900 US adults actually did across 68,879 unique Google searches collected in March 2025, of which 12,593 produced an AI summary. In that observed data, people clicked through to a traditional search result in just 8 percent of visits where an AI summary appeared, compared with 15 percent when it did not. About 18 percent of all searches in the sample produced a summary at all. This is measured behavior, not a stated intention, and it is the single strongest piece of evidence in this entire study. It tells you plainly that the click, the thing every small business website has been built and optimized to earn for two decades, is already worth roughly half what it used to be worth the moment an AI answer sits above it.
In Pew's own measured search sessions, an AI summary did not just add information to the page. It removed roughly half the reason to click anywhere else.
How Fast, and How Far the Surface Is Moving
The rollout itself is well documented. Google introduced AI Overviews in the US in May 2024, expanded to more than 100 countries by October 2024, and by May 2025 had the feature live in over 200 countries and territories and more than 40 languages, essentially everywhere except Cuba, Iran, and mainland China. In 2026 Google went a step further and put its position in writing for the first time, stating in its own published guidance that optimizing for generative AI search is, in its own words, still SEO. That sentence matters because it is Google telling the market, on the record, that this is not a separate discipline to be handled later.
How much of current search volume the feature actually touches is a fuzzier number. A Wikipedia compilation citing an unnamed third party SERP tracker puts the March 2026 figure at roughly 48 percent of Google queries showing an AI Overview, up from about 6.49 percent a year earlier. We are treating that specific percentage as directional, not confirmed, because neither the tracking methodology nor the vendor behind it is named in the source. The trajectory it describes, a feature going from a small minority of searches to a large share of them inside about a year, is consistent with everything else in this evidence set even if the exact number needs a named, corroborated tracker before it should be quoted as fact.
When the Answer Occupies the Screen
Even where the click still happens, it is competing for less visible space. A Botify and Demandsphere study found that when an AI Overview and a featured snippet appear together on a results page, they can occupy roughly 67.1 percent of visible screen space on desktop and 75.7 percent on mobile. That means a business ranking first organically can still be pushed below the fold before a visitor ever scrolls. We are flagging this as a single vendor study, not an industry consensus, but it lines up directionally with the click-through drop Pew measured independently.
A second, separate signal points the same way from a different angle. Penske Media Corporation's September 2025 lawsuit against Google alleges AI Overviews already appear on 20 percent of the Google searches that link to its properties, including Rolling Stone and The Hollywood Reporter, a share the filing expects to keep rising and blames for reduced referral traffic that funds original journalism. A litigation allegation is not audited data and we treat it accordingly, but it is a large publisher stating under legal exposure that the pattern Pew measured in aggregate is showing up in its own analytics too.
The Divide That Never Closed
Before any second, machine-legibility divide can be taken seriously, it is worth remembering the first one is still open. U.S. Census Bureau data cited in 2026 puts the number of Americans who still lack access to high-speed broadband at over 15.7 million. This is not a historical footnote from the dial-up era; it is a current figure. Any conversation about whether small businesses are legible to AI answer engines has to sit on top of the fact that a meaningful slice of the country still does not have reliable broadband in the first place. If a Digital Divide 2.0 exists, it is not replacing the original divide, it is stacking on top of a gap that thirty years of infrastructure investment has not fully closed.
Machine-Legibility: What the Web Actually Looks Like Today
Structured data, the markup that tells a machine what a page is actually about rather than making a human infer it, is now closer to a majority pattern than a niche practice. W3Techs reports that 79.4 percent of all websites use at least one structured data format, with JSON-LD specifically used by 54.7 percent of sites, microdata by 21.9 percent, Open Graph by 71.5 percent, and generic RDFa by 37.5 percent (formats are not mutually exclusive, so these overlap).
The caveat matters as much as the number. This is an all web average, and it is almost certainly dominated by content management systems and platforms, WordPress, Shopify, and similar tools, that auto inject basic schema whether the site owner ever thinks about it or not. It tells us structured data is now the default posture of the web as a whole. It tells us nothing about whether the average small, owner-operated business site, especially one hand-built or on an older platform without that auto injection, is keeping pace with that default. No public source we could locate breaks this figure out by business size, which is precisely the gap this study cannot close on its own.
Who Gets Cited, and From Where
If AI answers cite sources rather than simply summarizing the open web at random, the identity of those sources matters more than any individual business's on-page optimization. A Semrush analysis of 200,000 US keywords that triggered a Google AI Overview during a September 2024 window found the most frequently cited source across that sample was Quora, with Reddit second. Large user-generated content platforms, not individual business websites, dominated the citations in that data, regardless of how technically well built any given small business's own site was. This is a single vendor study and should be read as such, but it is the clearest available signal that winning a citation inside an AI answer may depend as much on your presence and reputation on the platforms an engine already trusts as it does on your own site's schema markup.
Review signals reinforce the same picture from the buyer side. BrightLocal's 2026 consumer survey found 97 percent of US consumers read online reviews when researching a local business, 41 percent now say they always read reviews, up from 29 percent a year earlier, and 68 percent require a minimum four-star rating before they will consider a business at all. A separate BrightLocal survey of local search practitioners found 93 percent had experimented with generative AI tools and 92 percent specifically with ChatGPT, while schema markup services were offered by 63 percent of agencies versus 50 percent of independent freelancers and consultants, a gap that hints at a resourcing divide inside the industry meant to serve small businesses. Both BrightLocal figures come from a single commercial vendor with a stake in the local SEO market, so we grade them emerging rather than established, useful as directional evidence, not as settled fact.
Large user-generated content platforms dominated the citations in the sampled data, regardless of how technically well built any individual business's own site was.
The Gap Nobody Has Measured Yet
Small and medium enterprises are not a niche concern in this conversation, they are most of the economy. World Bank, OECD, and European Union consensus statistics put SMEs at roughly 90 percent of all businesses worldwide and more than half of global employment, rising to 99 percent of all businesses inside the European Union specifically. If a machine-legibility gap between small and large firms is real, it would touch the large majority of the world's commercial activity, not a fringe segment.
And yet, after searching academic databases, market reports, and government sources, we could not locate a single public study that directly measures a firm-size-segmented machine-legibility gap, meaning nobody has published data comparing how well small business websites versus large competitor websites are structured for a machine to read, nor a firm-size breakdown of who actually gets cited inside AI answer engines. The W3Techs structured data figure is an all web average that cannot be disaggregated by business size with the sources available to us. The OECD's own SME digitalization research and the European Commission's Digital Economy and Society Index almost certainly contain relevant data, but their pages were not accessible to us this session, which is a retrieval limitation on our end, not evidence the data does not exist, and it needs a direct follow up pass before publication of any firm claim either way. Until that gap is closed, by us or by someone else, treat any vendor's claim to have already proven small businesses lag large ones in AI answer citation as unsubstantiated. It is a reasonable hypothesis given everything else in this evidence set. It is not yet a measured fact.
The evidence, in numbers
Key findings, dated and sourced
-
Google users who encountered an AI Overview clicked through to a traditional search result in only 8% of visits, versus 15% when no AI summary appeared; about 18% of searches in the sample produced an AI summary.
established Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results (Athena Chapekis, Anna Lieb); 900 US adults, 68,879 unique Google searches, 12,593 with AI summaries, collected March 2025 (2025-07-22)
-
Google AI Overviews appear on roughly 48% of total Google Search queries as of March 2026, up from about 6.49% a year earlier. The tracking methodology and vendor are unnamed, so this is directional rather than confirmed.
contested Wikipedia (AI Overviews article), compiling unnamed third party SERP tracking data, Wikipedia: AI Overviews (2026-03)
-
AI Overviews launched in the US in May 2024, reached 100+ countries by October 2024, and by May 2025 were live in over 200 countries and territories and 40+ languages, essentially everywhere except Cuba, Iran, and mainland China.
established Google (product history, via Wikipedia), Wikipedia: AI Overviews (2025-05)
-
When AI Overviews and featured snippets appear together, they occupy roughly 67.1% of visible screen space on desktop and 75.7% on mobile, meaning even a first place organic ranking can sit below the fold.
contested Botify x Demandsphere SERP real estate study, cited in Wikipedia: AI Overviews (2024-12)
-
79.4% of all websites use at least one structured data markup format; JSON-LD is used by 54.7% of sites, microdata by 21.9%, Open Graph by 71.5%, and generic RDFa by 37.5% (formats are not mutually exclusive).
established W3Techs: Usage statistics of structured data formats for websites (2026-07)
-
Penske Media Corporation's lawsuit against Google alleges AI Overviews already appear on 20% of Google searches that link to its properties, a share it expects to keep rising. This is a litigation allegation, not independently audited data.
contested Penske Media Corporation (litigation filing), Penske Media Corp. v. Google lawsuit, cited in Wikipedia: AI Overviews (2025-09)
-
97% of US consumers read online reviews when researching local businesses; 41% now say they always read reviews, up from 29% a year earlier; 68% require a minimum four-star rating. A single vendor commercial panel survey.
emerging BrightLocal Local Consumer Review Survey 2026 (n=1,002 US adults, SurveyMonkey panel) (2026-02-11)
-
Among local search marketing practitioners, 93% had experimented with generative AI tools and 92% with ChatGPT specifically; 46% believed local SEO would get harder over the following year; schema markup services were offered by 63% of agencies versus 50% of independent freelancers and consultants.
emerging BrightLocal Local Search Industry Survey 2023 (n=534, 72% US-based) (2023-08-17)
-
Weekly use of AI chatbots for news stands at 7% globally, rising to 15% among people under 25; ChatGPT specifically is used weekly for news by about 4% of respondents on average across markets; search engines are used weekly to check or verify news by 33%. Audiences expect AI to make news less trustworthy (net score -18) and less accurate (net score -8).
established Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025 (48-market survey) (2025-06-17)
-
6.04 billion people use the internet globally (73.2% penetration); 5.66 billion hold active social media identities; more than 1 billion people use LLM or generative AI platforms monthly; ChatGPT has 550 million monthly mobile app users and roughly 500 million unique web devices.
established Meltwater / We Are Social (Kepios analysis), Digital 2026 Global Overview Report (2025-10)
-
Google holds 91.27% of global search engine market share as of June 2026; Bing holds 4.68%; all other engines combined hold under 4%.
established StatCounter Global Stats, StatCounter: Search Engine Market Share Worldwide (2026-06)
-
SMEs make up roughly 90% of all businesses worldwide and account for more than half of global employment; in the European Union specifically, 99% of all businesses are classified as SMEs.
established World Bank / OECD / European Union (consensus definitional statistics), Wikipedia: Small and medium enterprises (2026)
-
In 2026 Google published official documentation stating that optimizing for generative AI search is, in its own words, still SEO, the first time it formally documented this position outside of conference talks.
established Wikipedia: Generative engine optimization, citing Google Search Central documentation (2026)
-
Over 15.7 million Americans still lack access to high-speed broadband, showing the original infrastructural digital divide remains unresolved even as a second, machine-legibility layer of disadvantage emerges on top of it.
established U.S. Census Bureau, Wikipedia: Digital divide, citing 2026 US Census Bureau data
-
Analyzing 200,000 US keywords that triggered a Google AI Overview during a September 2024 window, the most frequently cited source was Quora, followed by Reddit; large user-generated content platforms, not individual business websites, dominated the sample's AI answer citations.
contested Semrush AI Overviews source citation study, cited in Wikipedia: AI Overviews (2025-06 (study); published 2025-07-22)
Learning outcomes
What this study teaches
- Assume less of your buyer's attention is reaching your website directly than it used to. Pew's measured search data shows the click-through rate to a traditional result falls from 15% to 8% the moment an AI summary appears on the page.
- Structured data is now closer to a majority practice than a niche one across the general web (54.7% of sites use JSON-LD specifically), so leaving your own site without it increasingly reads as invisible infrastructure rather than a neutral choice.
- Be skeptical of any vendor claiming to have already proven that small businesses lag large ones in AI answer citations. No public study we could find measures that gap directly. It is a reasonable hypothesis, not yet a fact.
- Third party platforms already dominate what gets cited inside AI answers in the available sampled data (Quora and Reddit led one analysis), so your presence and reputation on the platforms an engine already trusts may matter as much as your own site's technical polish.
- The original broadband access gap has not closed (15.7 million Americans still lack high-speed access), so any new machine-legibility gap is stacking on top of an older, unresolved one, not replacing it.
Honest limits
What this does not yet settle
- No public study located this session directly measures a firm-size-segmented machine-legibility gap, meaning structured data and schema adoption compared between small and large businesses specifically. The W3Techs figure is an all web average almost certainly dominated by CMS and platform defaults, and it cannot be broken out by business size with the sources available to us. This is the central gap the whole thesis depends on closing.
- No independent, methodologically transparent census of which businesses, by size, actually get cited inside AI answer engines exists publicly. The available studies work at the platform level (screen occupation) or the source type level (Quora and Reddit dominance), not the individual business level.
- OECD SME digitalization reporting and the European Union's Digital Economy and Society Index likely contain exactly this kind of small versus large digital adoption data, but their pages returned access errors during this research pass. That is a retrieval limitation, not evidence the data does not exist, and it needs a direct follow up before any claim from those sources is published.
- US Census Bureau and Federal Reserve small business surveys likely track technology adoption by employee size bracket in ways relevant to this question, but the specific current pages could not be retrieved this session and need direct follow up.
- The BrightLocal figures measure marketing practitioner behavior and consumer review habits, both useful, but neither is a direct technical measurement of small business website readiness, and both come from a single commercial vendor with a stated interest in the local SEO market.
- The AI Overview prevalence trend (48% versus 6.49%) is sourced through a Wikipedia compilation without a named, corroborated primary tracker and should not be quoted as a confirmed figure until a named vendor's methodology can be checked directly.
- No independent census of the AI answer layer exists yet at all. Until one exists, any claim about a firm-size citation gap rests on inference from adjacent findings, not on direct measurement, and closing that gap is the logical next step for this line of research.
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
Is buyer attention really moving from links to synthesized summaries, or is that just a theory?
It is measured, not theoretical. Pew Research Center watched real Google search sessions, 900 US adults across 68,879 unique queries collected in March 2025, and found people clicked through to a traditional search result in only 8 percent of visits when an AI summary appeared, versus 15 percent when it did not. About 18 percent of searches in that sample produced a summary at all. That is the study's most solidly established finding.
Are small businesses actually losing ground to larger competitors inside synthesized answers?
That specific claim has not been measured by any public source the study could locate, so it treats it as an open question rather than a settled fact. What is known is that structured data, the markup that helps a machine understand a page, is now used by 79.4 percent of all websites, with JSON-LD specifically at 54.7 percent, but that figure is an all-web average likely propped up by CMS platforms auto-injecting schema, and it cannot be broken down by business size with the sources available. Be skeptical of any vendor who claims to have already proven small businesses lag large ones here.
How much of Google search now shows an AI Overview instead of regular links?
A Wikipedia compilation citing an unnamed third-party tracker puts it at roughly 48 percent of Google queries in March 2026, up from about 6.49 percent a year earlier. The study flags this specific number as directional rather than confirmed because the tracking methodology and vendor are not disclosed, though the overall direction of growth is not in doubt. What is documented on the record is the rollout itself: AI Overviews launched in the US in May 2024, reached over 100 countries by October 2024, and by May 2025 were live in more than 200 countries and territories in 40-plus languages.
If a business ranks first organically, does that still guarantee visibility on the results page?
Not necessarily. A Botify and Demandsphere study found that when an AI Overview and a featured snippet appear together, they can occupy roughly 67.1 percent of visible screen space on desktop and 75.7 percent on mobile, which means even a first-place organic result can sit below the fold before anyone scrolls. The study flags this as a single vendor study rather than an industry consensus, but notes it lines up directionally with Pew's independently measured click-through drop.
What should a small business actually do with all this, given how much is still unproven?
Assume less of your buyer's attention is reaching your website directly than it used to, since the click-through drop from 15 percent to 8 percent is well established. Beyond that, the study's most honest advice is to focus on where your market's attention actually sits rather than only your own site: a Semrush analysis of 200,000 keywords that triggered an AI Overview found Quora and Reddit were the most frequently cited sources, ahead of individual business sites, so your presence and reputation on platforms an engine already trusts may matter as much as your own site's technical polish. And remember the older divide never closed, since over 15.7 million Americans still lack high-speed broadband, so any new machine-legibility gap would be stacking on top of that, not replacing it.
Provenance
References
- Chapekis, A. and Lieb, A., "Google users are less likely to click on links when an AI summary appears in the results," Pew Research Center, July 22, 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/
- "AI Overviews," Wikipedia, accessed 2026 (compiling SERP tracking, Botify x Demandsphere, and Penske Media litigation data). https://en.wikipedia.org/wiki/AI_Overviews
- "Usage statistics of structured data formats for websites," W3Techs, July 2026. https://w3techs.com/technologies/overview/structured_data
- "Local Consumer Review Survey 2026," BrightLocal, February 11, 2026. https://www.brightlocal.com/research/local-consumer-review-survey/
- "Local Search Industry Survey 2023," BrightLocal, August 17, 2023. https://www.brightlocal.com/research/local-seo-industry-survey/
- "Digital News Report 2025," Reuters Institute for the Study of Journalism, University of Oxford, June 17, 2025. https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2025
- "Digital 2026 Global Overview Report," Meltwater / We Are Social (Kepios analysis), via DataReportal, October 2025. https://datareportal.com/reports/digital-2026-global-overview-report
- "Search Engine Market Share Worldwide," StatCounter Global Stats, June 2026. https://gs.statcounter.com/search-engine-market-share
- "Small and medium enterprises," Wikipedia, accessed 2026. https://en.wikipedia.org/wiki/Small_and_medium_enterprises
- "Generative engine optimization," Wikipedia, citing Google Search Central documentation, accessed 2026. https://en.wikipedia.org/wiki/Generative_engine_optimization
- "Digital divide," Wikipedia, citing 2026 U.S. Census Bureau data. https://en.wikipedia.org/wiki/Digital_divide
- "AI Overviews source citation study," Semrush, published July 22, 2025 (study window September 2024). https://www.semrush.com/blog/ai-overviews-study/
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.