“Welp, that happened faster than I predicted. We are clearly on the other side now.” That is the CEO and co-founder of Cloudflare — the company that processes more web traffic data than almost any other organisation on earth — reacting to his own company’s finding that bot traffic now represents 57.5% of all HTTP requests on the internet. He expected this crossover in 2027. It happened in June 2026. The internet just changed hands.”
Ground Truth · Episode 13 · June 25 2026For the first two decades of the commercial internet, the fundamental assumption underlying every business model built on web traffic was that the traffic consisted primarily of humans. Advertisers bought impressions because humans saw them. Publishers built audiences because humans read their content. Search engines ranked pages because humans found them useful. E-commerce sites optimised conversion funnels because humans moved through them. The entire economic architecture of the modern internet was built on human attention as the scarce resource being competed for. In June 2026, according to data published by Cloudflare this week, that assumption is no longer accurate. 57.5% of all HTTP web requests are now machine traffic — AI crawlers, automated scrapers, agentic browsing systems, content ingestion pipelines, and the vast ecosystem of bots that AI companies and their customers deploy to consume the web at computational speed. Cloudflare’s CEO Matthew Prince said he didn’t expect this crossover until at least 2027. It happened in June 2026. The internet just changed hands. Welcome to Episode 13 of Ground Truth.
Not All Bots Are the Same. But the Aggregate Is Now the Majority.
The 57.5% figure requires context. Bot traffic has existed for as long as the web — search engine crawlers, monitoring tools, security scanners, and content aggregators have always been part of the traffic mix. What changed in 2025 and accelerated through 2026 is the composition and volume of bot traffic. The new majority is driven primarily by AI-related traffic: the crawlers deployed by AI companies to ingest web content for model training (ClaudeBot, GPTBot, Google-Extended, and their many variants), the agentic browsing systems that AI assistants use to research queries on behalf of users, the automated pipelines that enterprise customers use to feed real-time web data into their AI workflows, and the increasingly sophisticated AI-generated content republication bots that are flooding the web with machine-produced pages designed to attract both human and machine attention.
The asymmetry within this traffic is the detail that matters most for publishers. As covered in the crawler asymmetry data from Episode 07, ClaudeBot was crawling 11,122 pages for every single human visit Anthropic sent back to publishers. GPTBot was at 857:1. These ratios represent the economic reality of the machine web: AI companies are consuming web content at enormous scale and returning almost none of the traffic value that consumption represents in the human-traffic model. The content being consumed is the product of significant human labour — journalism, research, creative writing, professional expertise — and the economic model that funded that production was built on human visitors who saw advertising, subscribed to publications, and clicked through from search results. At 57.5% bot traffic, that model is structurally broken and the political economy of the web is being renegotiated in real time.
June 2026: 57.5% bot traffic, 42.5% human traffic (Cloudflare). Cloudflare CEO’s prediction: crossover by 2027. Actual crossover: June 2026. 1 year early. ClaudeBot crawl ratio: 11,122 pages crawled per human visit returned. GPTBot: 857:1. Perplexity: 190:1 (up from 95:1). Google Googlebot: 5:1. Google AI Mode: 1 billion monthly active users. Average AI Mode query: 3x longer than traditional search. Planning queries growing 80% faster than overall AI Mode growth. Image searches: +40% month-over-month. Human web traffic share: declining. Machine web traffic share: majority, and growing.
Google AI Mode Crossed a Billion Users. Publishers Are Not Celebrating.
Google confirmed this week that its AI Mode — the search interface that answers queries directly rather than sending users to external websites — has surpassed one billion monthly active users globally. Sundar Pichai called the AI transition a seamless “continuum” from classic search and noted that AI Mode is driving search queries to an all-time high. The queries are longer, more complex, and more planning-oriented than traditional searches. This is, on the metrics Google reports, a success story.
Publishers are experiencing it differently. The phenomenon that industry analysts are now calling “Google Zero” describes the outcome that increasingly defines the AI Mode experience: Google provides attribution — a source name, sometimes a link — but the user’s question is answered directly on the search page. There is no click. There is no traffic. The publisher’s content has been used to construct the answer and the publisher receives the appearance of attribution without the economic value that attribution was supposed to represent. Pichai’s assurance that “sources and links will always remain part of AI Mode” addresses the visibility concern but not the traffic concern. Visible and visited are no longer the same thing. That distinction is the entire problem.
The economic implications are not abstract. Digital publishers have built their businesses on the assumption that search visibility translates into traffic, which translates into advertising revenue or subscription conversion. AI Mode preserves the first link in that chain — visibility — while breaking the second. Publishers who adapted to SEO over two decades are now facing a second structural disruption: the traffic that SEO was designed to generate is being intercepted by an AI answer layer that gives readers what they came for without sending them anywhere. Several major publishers — including news organisations, vertical content sites, and review platforms — have reported double-digit declines in organic search referral traffic in the first half of 2026. The decline is accelerating.
Google AI Mode has a billion users. The average query is three times longer. Planning queries are growing 80% faster than overall AI growth. By every metric Google measures, this is a success. By every metric publishers measure — referral traffic, advertising revenue, subscription conversion — this is a crisis. Both things are true simultaneously. The transition from search-as-discovery to AI-as-answer has winners and losers. Google is the winner. Publishers are the losers. The question is what comes next.Neal Lloyd · Ground Truth, Episode 13
New Reporting Names the Person Who Escalated the Vulnerability Report. It Is Jeff Bezos’s Lieutenant.
New reporting published this week by unrot.co and confirmed across multiple AI news trackers names Andy Jassy — CEO of Amazon and the man who reports directly to Jeff Bezos — as the person who escalated the Fable 5 vulnerability finding to the White House. The sequence, as now reported: Amazon researchers within AWS identified the Fable 5 vulnerabilities. Those findings were escalated internally at Amazon. Andy Jassy made the decision to report them to the White House directly rather than to Anthropic. The White House received the report from the CEO of Anthropic’s single largest investor and cloud provider. The recall followed.
The significance of this detail is not that Amazon acted in bad faith — there is no evidence of that, and responsible disclosure of genuine security vulnerabilities to appropriate authorities is, on its face, the correct response. The significance is what it reveals about the structural tensions inside the Anthropic-Amazon relationship. Amazon has invested $8 billion in Anthropic. It is Anthropic’s primary cloud infrastructure provider. It is simultaneously the organisation whose CEO made the decision to escalate a security finding about Anthropic’s most important product to the government — a decision that contributed to a recall that damaged Anthropic’s IPO narrative and enterprise relationships at the most sensitive commercial moment in the company’s history. The Jassy decision may have been entirely correct on the merits. It was also a decision that a partner-competitor made about a partner-investee in a way that had significant material consequences. The Anthropic S-1 will need to address this relationship with candour that goes beyond the standard “significant customer concentration” risk factor.
Meanwhile, Fable 5 and Mythos 5 remain offline today — June 25th — thirteen days after the recall. GPT-5.6 is appearing in ChatGPT Pro ahead of a scheduled late-June launch. The model that was pulled is offline. The competitor’s next model is already in preview. The market gap is accumulating every day the recall continues.
Three days ago, the White House issued a new executive order on AI and cybersecurity. The order creates an AI cybersecurity clearinghouse — a voluntary collaboration between the AI industry, critical infrastructure operators, and federal agencies — to coordinate vulnerability scanning, discovery, validation, and patch distribution. The order directs: NSA and CISA involvement in National Security Systems cybersecurity within 30 days. Treasury formation of an AI cybersecurity clearinghouse with voluntary industry participation within 30 days. OMB review of available federal grant funding for AI vulnerability detection within 30 days. The order explicitly frames AI capabilities as both a national strength and a new national security consideration requiring coordinated action. Read in the context of the Fable 5 recall, the White House order is not a coincidence. It is the institutional architecture being built around the policy stance that the recall enacted.
The Practical Implications of Living on the Other Side of the Bot Crossover
For publishers and content creators. The traffic model is broken. The question is not whether to adapt but how. The adaptations that appear most robust: direct audience relationships (newsletters, podcasts, community platforms) that do not depend on search referral traffic; paywalled content that AI crawlers cannot access and therefore cannot substitute; syndication deals with AI companies that convert the crawl-without-return dynamic into a licensed relationship with economic terms; and product and service businesses built on the credibility that content generates, rather than businesses that monetise the content itself through advertising. The New York Times’s landmark lawsuit against OpenAI — and the FAIR News Act now signed in New York — represent the legislative pathway. Whether those mechanisms produce adequate compensation on a timescale that saves existing publishers is genuinely uncertain.
For advertisers. The 42.5% of traffic that is human is increasingly self-selected — people who chose to navigate directly to a site, or to click through from a non-AI referral. That audience is smaller, more intentional, and potentially more valuable per person than the broad search-referred traffic of the pre-AI era. Advertising in a world of 42.5% human traffic looks less like reaching mass audiences through search and more like reaching concentrated, high-intent communities through direct relationships. The advertising models that work in that environment — sponsorships, native integration, community-based formats — are different from the display advertising optimised for high-volume search traffic. The entire adtech stack is repricing.
For AI companies. The crawl-heavy, return-nothing dynamic documented in Episodes 07 and this episode is a reputational and legal liability that is accumulating faster than the industry’s responses to it. The FAIR News Act in New York, the ongoing publisher lawsuits, the EU’s AI Act content provisions, and the growing political appetite for publisher compensation frameworks all reflect the same underlying dynamic: the current arrangement, in which AI companies consume web content at enormous scale and return almost no traffic value, is politically and legally unsustainable. The AI companies that resolve this proactively — through licensing frameworks, revenue sharing, or traffic commitments — will face less regulatory friction than those that wait for legislation. The clock is running.
For the open web. The most serious long-term risk of the bot-majority internet is the erosion of the content ecosystem that the bots are consuming. If AI systems make it economically unviable to produce high-quality original content for the web — because the content is consumed by AI without generating the traffic revenue that funded its production — the quality and diversity of web content will decline. AI systems trained on declining-quality web content will produce worse outputs. Worse AI outputs reduce the economic value of AI. The feedback loop runs in reverse. The open web and the AI ecosystem are, in this sense, mutually dependent in a way that the current crawl-without-return dynamic is not serving.
The internet was built on human attention as the scarce resource. Bot traffic is now the majority. The economic architecture built on human attention — advertising, search referral, content monetisation — is being restructured by a technology that consumes the product of that architecture without participating in its economics. The adjustment will be painful, contested, and necessary. It is happening right now.Neal Lloyd · Ground Truth, Episode 13
Ground Truth, Episode 13 · June 25 2026
Neal Lloyd covers the real-world impact of AI — money, power, geopolitics, and the stories behind the headlines. Ground Truth is his daily AI news and analysis series on emdexter.blogspot.com.
- Ep 01The Gold Rush
- Ep 02ChatGPT Knows Everything
- Ep 03Siri Is Now Google
- Ep 04America’s AI Law Is a Mess
- Ep 05Is AI Taking Your Job?
- Ep 06Microsoft vs Everyone
- Ep 07SpaceX Is Trading
- Ep 08The Government Pulled Fable 5
- Ep 09Trump and Bernie Want to Own AI
- Ep 10SpaceX Buys Cursor for $60B
- Ep 11The Fable 5 Truth
- Ep 12The Week That Changed Everything
- Ep 13Bots Now Outnumber HumansYou are here
