Global & MSME Commerce
The Platform Tax and Disintermediation
AI Overviews already send far fewer clicks to source sites than a normal results page does, but nobody, not even the platforms themselves, has yet measured whether marketplaces, OTAs, and lead-generation businesses are winning or losing the answer layer built to replace that click.
Part of MSME & Global Commerce in the Insights library.
Abstract
When Google shows a synthesized answer instead of a plain results page, Pew Research found people click through to an outside source only 8% of the time, versus 15% when no summary appears, and click a source cited inside the summary itself in just 1% of visits. Publishers and at least one lead-gen-adjacent intermediary are already litigating and restructuring in response, and regulators on both sides of the Atlantic are testing the fee side and the answer side of the squeeze at once. What the public record does not yet contain is any independent, vertical-specific number showing whether marketplaces, OTAs, or lead-generation platforms specifically are gaining or losing ground in the AI-answer layer.
The surface where discovery happens is visibly changing shape, in the United States and unevenly around the world. Pew Research Center's tracking of 900 U.S. adults across nearly 69,000 real Google searches found that when a synthesized AI summary appears on the results page, people click through to an outside link only 8% of the time, compared with 15% when no summary appears, and click a source cited inside the summary itself in just 1% of visits (established, July 2025). A single-vendor study from Botify and Demandsphere, which we tier contested pending independent corroboration, put the combined footprint of AI Overviews and featured snippets at roughly two-thirds of the visible screen on desktop and just over three-quarters on mobile. Regulators are moving too, in opposite directions: the U.S. Eighth Circuit vacated the FTC's click-to-cancel rule in July 2025 on procedural grounds, loosening one federal check on subscription and cancellation friction, while the European Commission opened an investigation in December 2025 into whether Google's AI Overviews themselves breach EU competition law by foreclosing traffic to smaller publishers, a finding we tier emerging since we could not locate the Commission's own primary announcement this session. None of this is settled. What is established is that the shape of the page itself, how much of it an answer occupies and how often a person still leaves it, is moving in one direction.
For the person doing the searching, the practical effect is that the answer is increasingly the destination rather than a doorway. Reuters Institute's Digital News Report 2025 found that, globally, 7% of people use AI chatbots weekly for news, rising to 15% among under-25s and 18% in India, the highest of the markets surveyed, against just 3% in the UK, the lowest (established). The same report found that when people are checking whether something might be false, younger respondents lean proportionally more on AI chatbots and social platforms, while older users still default to established news brands and official sources, a generational split that will not close on its own. Layered on top of that is a citation pattern worth sitting with: a Semrush study, contested because it rests on a single vendor, found the domains most frequently cited inside Google AI Overviews are Quora and Reddit, forum and community platforms, not brand-owned commercial or marketplace domains. Put together, these findings describe a buyer who increasingly forms a first impression, and sometimes a final one, from a synthesized answer built substantially on what strangers said in a forum thread, not from the marketplace listing, the OTA page, or the lead-gen form that used to be the first and only stop.
For a marketplace, an OTA, or a lead-generation platform, this redraws the map of where the return on visibility actually sits, and the public evidence shows that nobody has drawn that map yet for these specific categories. We could not find a verified take-rate trend for Booking Holdings or Expedia, a verified effective take rate for Amazon, or any vertical data on lead-gen platforms like Zillow Premier Agent or Angi under AI-answer pressure; every figure in the record so far describes general search behavior, publisher-side stress, or regulatory posture, not a specific vertical's share of the answer. Some intermediaries are visibly reacting: Chegg filed suit against Alphabet in February 2025 and has since said publicly it is exploring a sale or take-private transaction, an education- and lead-gen-adjacent business restructuring in real time (emerging). Others are betting on the opposite move, trying to become the answer and transaction layer itself; Perplexity AI reached roughly a $20 billion valuation by September 2025 while pursuing agentic shopping ambitions, even as Wired and Cloudflare separately documented it using undisclosed, spoofed crawlers to scrape sites that explicitly block scraping (established, with real reputational cost attached). Until a business can see, category by category, where its own market's attention actually sits today, in the map pack, on a forum thread, inside a chatbot's answer, or nowhere an engine currently looks, it is guessing at which side of this redrawing it is on. That is precisely the map a Visibility Corpus is built to produce.
The data, in one read
The Click That Used to Fund the Toll Booth
Marketplaces, OTAs, and lead-generation platforms have always made their money on the same basic mechanic: a buyer searches, lands on a results page, clicks through, and the platform charges someone (the seller, the hotel, the agent) a fee for having stood in the middle of that click. Call it the platform tax. It only works if the click keeps happening.
Pew Research Center's tracking of 900 U.S. adults across nearly 69,000 real Google searches found that when a synthesized answer appears on the results page, people click through to an outside link only 8% of the time, compared with 15% when no summary appears. Roughly half the click, in one of the cleanest independent behavioral panels available. And of the people who do interact with the summary, only 1% click a source cited inside it. The answer, in other words, is increasingly treated as the finish line, not a signpost toward one.
A separate, single-vendor study from Botify and Demandsphere, which we tier contested until it's independently corroborated, estimated that AI Overviews and featured snippets together occupy roughly 67.1% of the visible screen on desktop and 75.7% on mobile when both appear. Even without treating that figure as settled, it points the same direction as Pew's harder number: there is simply less room, and less reason, left on the page for the old toll booth to sit in.
When a synthesized answer appears on the results page, people click a source cited inside it in just 1 out of every 100 visits.
Who the Machine Credits When It Answers
If fewer people are clicking out, the next question is who still gets credit when the answer names a source at all. A Semrush study, which we tier contested because it rests on a single vendor without independent replication, found that the most frequently cited domain inside Google AI Overviews is Quora, followed by Reddit. Not a brand site. Not a marketplace listing. A forum.
That is a meaningful signal even at contested tier, because it describes a structural pattern rather than a one-off number: right now, the answer layer appears to trust what strangers said about a product, a hotel, or a service more than it trusts the commercial page selling it. For a marketplace or OTA whose entire business model depends on being the authoritative, transactable source, that is worth watching closely, and worth verifying directly for your own category rather than assuming it holds everywhere.
The Threat Is Not Landing Evenly
It would be a mistake to treat this as one uniform wave hitting every market and every age group at the same speed. Reuters Institute's Digital News Report 2025 found that, globally, 7% of people use AI chatbots weekly for news, but that rises to 15% among under-25s and to 18% in India, the highest of the markets surveyed, while sitting at just 3% in the UK, the lowest.
The same report found a related generational split in trust: when checking whether something might be false, younger respondents lean proportionally more on AI chatbots and social platforms, while older users still default to established news brands and official sources. For any business selling across age groups or geographies, that means the disintermediation risk in this study is not a single dial. It is a set of dials moving at different speeds depending on who your buyer is and where they live, and averaging them into one national number would hide exactly the detail that matters.
18% of people in India use AI chatbots weekly for news. In the UK, it's 3%. The same shift is not happening at the same speed everywhere.
Stress Already Showing Up in Filings and Newsrooms
This is not only a theoretical pressure. Chegg, an education and lead-gen-adjacent business, filed suit against Alphabet and Google in February 2025 alleging AI Overviews push students toward low-quality, unverified summaries in violation of antitrust law, and has since said publicly it is exploring a sale or take-private transaction, an observable case of a real business restructuring under this exact pressure (emerging tier).
Penske Media, publisher of Rolling Stone and The Hollywood Reporter, sued Google alleging AI Overviews illegally reproduce its content and reduce site visits, with the complaint claiming AI Overviews appear on 20% of searches linking to Penske-owned sites, a figure Google disputes, saying its summaries send traffic to a greater diversity of sites. Both statements come from an active, unresolved lawsuit, so we tier the claim contested rather than settled fact. A Munich Regional Court went further, issuing a temporary injunction against Google after two publishers alleged AI Overviews falsely characterized their businesses as running scams, and holding that, unlike ordinary search results that merely link out, AI summaries generate independent, substantive statements that Google can be held liable for. That, too, is a single first-instance ruling and we tier it contested, but the legal reasoning itself is notable.
None of this is confined to publishers filing lawsuits. Industry insiders have coined the term "Google Zero" for the point at which search stops sending any click traffic at all, and outlets including The Verge and Wired have cited it publicly as the reason they're pivoting toward direct-reader relationships such as newsletters and membership, with Wired's editor-in-chief describing the shift in stark terms. That term is an industry framing more than an independently measured trend, worth taking seriously as a signal, not as a settled statistic.
Wired's editor-in-chief has called the shift a "traffic apocalypse."
Regulators Pulling From Both Directions at Once
Two regulatory threads are moving in this study at the same time, and they pull on intermediaries from opposite ends. On the fee side, the U.S. Court of Appeals for the Eighth Circuit vacated the FTC's click-to-cancel rule in July 2025 on procedural grounds, which would have required subscription and negative-option sellers to make cancellation as easy as sign-up. The FTC still holds other statutory tools, but that specific rule is gone for now, an established fact that loosens one check on subscription-style monetization.
On the answer side, the European Commission announced in December 2025 that it is investigating whether Google's AI Overviews themselves breach EU competition law, centered on concerns that they foreclose traffic and visibility to smaller publishers. We tier this emerging because we could not locate the Commission's own primary announcement this session, only secondary reporting of it. Read together, one thread is loosening pressure on how intermediaries collect their fee, and the other is applying new pressure on the platforms that may be starving them of the traffic that fee depends on. Neither is resolved, and both are worth tracking rather than assuming.
The Counter-Bet: Become the Answer Instead of Losing to It
Not every player is playing defense. Perplexity AI reached a valuation of roughly $20 billion by September 2025, with a later round reportedly valuing it near $21.2 billion in early 2026, while pursuing agentic shopping and transaction ambitions that would place it inside the discovery-to-purchase path rather than downstream of it.
That ambition comes with real friction attached. Perplexity has faced litigation and public disputes over content use with the BBC, Dow Jones, the New York Post, and the New York Times, and independent technical analyses by Wired and Cloudflare separately found it using undisclosed web crawlers with spoofed identifying information to scrape sites that explicitly block scraping. The valuation and the litigation are both established facts, sitting side by side, and that combination is worth naming plainly: the re-intermediation bet is real money and real ambition, built on methods that have already drawn scrutiny from the publishers whose content it depends on.
The Number Nobody Has Published Yet
Everything above describes general search behavior, publisher-side stress, and regulatory posture. None of it is specific to marketplaces, OTAs, or lead-generation platforms as categories. We looked for a verified take-rate trend for Booking Holdings or Expedia, a verified effective take rate for Amazon, and vertical data on lead-gen businesses like Zillow Premier Agent or Angi under AI-answer pressure, and found none that met the bar of a real, named, retrievable source.
We also excluded a widely circulated figure claiming AI Overviews appeared on more than 48% of Google queries by March 2026, because it traced back to an unnamed vendor through a secondhand citation with no retrievable primary report. It is not that the number is implausible. It is that it cannot be traced to a real source, and unsourced figures do not appear in this study.
Every click-through and citation-share number in this study, even the strongest ones, comes from a third-party research panel or an SEO-vendor estimate. No platform, Google, OpenAI, or Perplexity, has released official first-party data on how much traffic is actually being diverted. That is the state of the public record, and it is exactly the gap a direct, vertical-specific read of your own market is built to close.
No independent, vertical-specific share-of-answer number exists yet for marketplaces, OTAs, or lead-generation platforms. That is the gap this study leaves open.
The evidence, in numbers
Key findings, dated and sourced
-
When a Google AI Overview appears on a search results page, people click through to an external link only 8% of the time, versus 15% when no AI summary appears; they click a source cited inside the AI summary itself in just 1% of visits.
established Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results," based on tracking of 900 U.S. adults across 68,879 Google searches (12,593 with AI summaries), browsing data from March 2025 and search-result capture from April 7 to 17, 2025. (2025-07-22)
-
When AI Overviews and featured snippets both appear on a Google results page, together they occupy roughly 67.1% of the visible screen on desktop and 75.7% on mobile.
contested Botify x Demandsphere study; a single-vendor study not yet independently corroborated. (2024-12)
-
Globally, 7% of people use AI chatbots such as ChatGPT and Gemini weekly for news, rising to 15% among under-25s and 18% in India, the highest of the markets surveyed, versus 3% in the UK, the lowest. ChatGPT specifically is used weekly for news by 4% of respondents on average across markets. Publishers surveyed worry AI summaries could further reduce traffic to their sites and apps.
established Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025 (2025-06-17)
-
Among people checking whether information might be false, younger respondents rely proportionally more on AI chatbots and social media, while older users continue to favor established news brands and official sources.
established Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025 (2025-06-17)
-
The U.S. Court of Appeals for the Eighth Circuit vacated the FTC's click-to-cancel rule on procedural grounds, blocking enforcement of a rule that would have required subscription and negative-option sellers to make cancellation at least as easy as sign-up. The FTC retains other statutory tools to pursue problematic subscription practices.
established U.S. Court of Appeals for the Eighth Circuit; reported by National Law Review, "Eighth Circuit Vacates FTC's Click-to-Cancel Rule" (2025-07-08)
-
The European Commission announced in December 2025 that it is investigating whether Google's AI Overviews breach EU competition law, centered on concerns about foreclosure of traffic and visibility to content providers and smaller publishers.
emerging Secondary reporting; no primary European Commission press release was located this session, so treat as needing corroboration from europa.eu before further citation. (2025-12)
-
Penske Media Corporation, publisher of Rolling Stone and The Hollywood Reporter, sued Google alleging AI Overviews illegally reproduce its content and reduce site visits. The complaint alleges AI Overviews appear on 20% of searches linking to Penske-owned sites and that this share is expected to rise. Google publicly disputes traffic harm, stating AI Overviews send traffic to a greater diversity of sites.
contested Penske Media Corporation v. Google (active litigation); Google spokesperson, An allegation in an active, unresolved lawsuit, not a settled fact. (2025-09)
-
Chegg sued Alphabet and Google, filed February 24, 2025, alleging AI Overviews are causing students to shift toward low-quality, unverified AI summaries in violation of antitrust law. Chegg publicly stated it was exploring a sale or take-private transaction, an observable case of an education- and lead-gen-adjacent intermediary restructuring under AI-answer pressure.
emerging Chegg, Inc., Litigation filing. (2025-02-24)
-
A Munich Regional Court issued a temporary injunction against Google after two Munich-based publishers alleged AI Overviews falsely characterized their businesses as running scams and subscription traps. The court held that, unlike traditional search results that merely link out, AI Overviews generate independent, substantive statements for which Google can bear liability, and rejected Google's defense that users could verify claims via linked sources.
contested Munich Regional Court (Landgericht Munchen), A single first-instance ruling; exact month not specified in available reporting. (2026)
-
The most frequently cited source domain inside Google AI Overviews is Quora, followed by Reddit, meaning current AI-answer citation share is concentrated in user-generated forum platforms rather than brand-owned commercial or marketplace domains.
contested Semrush study; a single-vendor study not yet independently corroborated. (2025-06)
-
"Zero-click result" and "Google Zero," the point at which Google search stops sending any click traffic, are established, named industry concepts. Publishers including The Verge and Wired have publicly cited Google Zero as a reason for pivoting toward direct-reader relationships such as email newsletters and membership, with Wired's editor-in-chief describing the shift in stark terms. A related discipline, generative engine optimization, also called answer engine optimization or AI engine optimization, has emerged to optimize content for inclusion in the answers these engines produce, rather than in ranked listings.
emerging Nilay Patel (The Verge); Katie Drummond (Wired); industry-wide GEO/AEO terminology, Industry and trade-press usage of the terms. (2024-2025)
-
Perplexity AI reached a valuation of approximately $20 billion in September 2025, with a later funding round reportedly valuing it near $21.2 billion in early 2026. The company has faced litigation and public disputes over content use with the BBC, Dow Jones, the New York Post, and the New York Times, and independent technical analyses by Wired and Cloudflare separately found Perplexity used undisclosed web crawlers with spoofed user-agent strings to scrape sites that explicitly block scraping.
established Perplexity AI, Inc.; independently corroborated by Wired and Cloudflare, Valuation timeline, litigation history, and crawler findings. (2025-09)
Learning outcomes
What this study teaches
- A page ranking well no longer guarantees a click: independent data shows click-through roughly halves when a synthesized answer appears, so your funnel math needs to account for that directly, not assume it away.
- Being cited inside an AI answer is not the same as being clicked. Only 1% of visits click a source inside the summary itself, so citation and traffic are two different goals now.
- Right now, one vendor study found forum and community platforms capture more AI citation share than brand or marketplace pages do. Treat it as a signal worth checking against your own category, not a settled rule.
- The regulatory ground under both subscription cancellation and AI-answer competition claims is actively moving in the US and EU. Don't assume today's posture, on either side, holds for long.
- No public research yet tells marketplaces, OTAs, or lead-generation platforms specifically whether they're gaining or losing the AI-answer layer. That number has to be captured directly for your category, not borrowed from a general industry trend.
Honest limits
What this does not yet settle
- No verified take-rate or booking-channel-mix data for OTAs such as Booking Holdings or Expedia Group was retrievable this pass; that needs primary capture from filings or a licensed industry report.
- No verified effective take rate (referral fee plus fulfillment plus advertising as a share of sales) exists yet for Amazon or any other marketplace.
- No vertical data was found on lead-generation platforms such as Zillow Premier Agent, Angi, or LegalZoom under AI-answer pressure; this category is currently unevidenced beyond general market narrative.
- No independent, methodologically transparent share-of-answer measurement broken out by marketplace, OTA, or lead-generation vertical exists yet in the public literature. This is the central gap, and the AI-answer layer remains a measurement frontier, not a solved census.
- Every AI-Overview click-through and citation-share figure here comes from a third-party panel (Pew) or an SEO-vendor estimate (Semrush, Botify x Demandsphere); no platform has released official first-party click data, so the true size of the shift is estimated, not confirmed, by Google, OpenAI, or Perplexity themselves.
- Penske Media's 20% figure and Google's traffic-diversity counterclaim are both self-interested statements inside active, unresolved litigation and should not be read as settled facts.
- No evidence was found on whether marketplaces, OTAs, or lead-gen platforms are successfully re-intermediating by becoming citable or transactable inside AI-answer systems, beyond Perplexity's general ambitions. This is a fast-moving area needing dedicated follow-up.
- One widely circulated figure, that AI Overviews appeared on more than 48% of Google queries by March 2026, was excluded from this study entirely: it traces to an unnamed vendor through a secondhand citation with no retrievable primary report.
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
Do AI answers on Google really send fewer clicks to websites than a normal search results page?
Yes, and the drop is well-measured. Pew Research Center tracked 900 U.S. adults across nearly 69,000 real Google searches and found people click through to an outside link only 8% of the time when a synthesized AI summary appears, compared with 15% when no summary appears. Click-through roughly halves. Even fewer people, just 1% of visits, click a source cited inside the summary itself.
Are marketplaces, OTAs, and lead-generation platforms specifically losing traffic to AI answers?
That is the gap in the public record: nobody knows yet. The study looked for a verified take-rate trend for Booking Holdings or Expedia, an effective take rate for Amazon, and vertical data on lead-gen platforms like Zillow Premier Agent or Angi under AI-answer pressure, and found none that met the bar of a real, retrievable source. Every click-through and citation-share figure available today describes general search behavior or publisher-side stress, not a specific vertical's share of the answer.
Is this shift affecting every market and age group the same way?
No. Reuters Institute's Digital News Report 2025 found that globally 7% of people use AI chatbots weekly for news, but that rises to 15% among under-25s and 18% in India, the highest surveyed market, versus just 3% in the UK, the lowest. The same report found younger respondents lean more on AI chatbots and social platforms when checking whether something might be false, while older users still default to established news brands. A business selling across ages or geographies should treat this as several dials moving at different speeds, not one national trend.
Should I trust the claim that Quora and Reddit get cited more than brand or marketplace pages inside AI Overviews?
Treat it as a signal, not a settled fact. A Semrush study found the domains most frequently cited inside Google AI Overviews are Quora, then Reddit, rather than brand-owned commercial or marketplace domains, but the study rests on a single vendor without independent replication, so this analysis tiers it contested. It is worth checking against your own category directly rather than assuming it holds everywhere.
What should a marketplace, OTA, or lead-gen business actually do with this information?
Build your funnel math around the fact that a page ranking well no longer guarantees a click, since independent data shows click-through roughly halves once a synthesized answer appears, and recognize that being cited inside an AI answer is not the same as being clicked. Beyond that, the study is clear that no public research yet tells any of these three categories whether they are gaining or losing ground in the AI-answer layer specifically, so that number has to be captured directly for your own category rather than borrowed from a general industry trend.
Provenance
References
- Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results," 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/
- Reuters Institute for the Study of Journalism, University of Oxford, Digital News Report 2025, June 17, 2025. https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2025
- National Law Review, "Eighth Circuit Vacates FTC's Click-to-Cancel Rule," reporting on the U.S. Court of Appeals for the Eighth Circuit ruling, July 8, 2025. https://www.natlawreview.com/article/eighth-circuit-vacates-ftcs-click-cancel-rule
- Botify x Demandsphere study on AI Overview and featured-snippet screen occupancy, December 2024.
- European Commission investigation into Google AI Overviews and EU competition law, announced December 2025; corroboration from a primary europa.eu source is still needed.
- Penske Media Corporation v. Google, litigation complaint filed September 2025.
- Chegg, Inc. v. Alphabet Inc. and Google LLC, litigation filing, February 24, 2025.
- Munich Regional Court (Landgericht Munchen), temporary injunction ruling, 2026.
- Semrush study on AI Overview citation-source domains, June 2025.
- Nilay Patel (The Verge) and Katie Drummond (Wired) on "Google Zero" and generative engine optimization, 2024-2025.
- Perplexity AI, Inc.: valuation timeline, litigation history, and the Wired and Cloudflare crawler findings, 2025-2026.
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.