HUMAN Security Reports AI Traffic Surged 187% in 2025 as Bots Eclipsed Human Users

HUMAN Security's 2026 State of AI Traffic report, released Thursday, found that automated internet traffic grew nearly eight times faster than human activity in 2025. AI-driven traffic increased 187% from January to December, with traffic from AI agents and agentic browsers surging 7,851% year over year, according to the cybersecurity firm. The report, based on more than one quadrillion digital interactions processed through HUMAN's Defense Platform, confirms what Cloudflare CEO Matthew Prince warned at SXSW last week. An internet built for human users is turning into something else.

Automated traffic outran humans by a factor of eight

Automated traffic grew 23.51% year over year in 2025. Humans? Their traffic grew 3.10%. HUMAN Security attributes that widening gap to AI chatbots pulling people away from direct browsing. More and more, someone types a question into ChatGPT instead of opening a tab, and a bot goes out to fetch the answer. A person used to sit at the keyboard. Now a bot sits there on the person's behalf.

Ninety-five percent of AI-driven automation wound up in retail, streaming, and travel, the three sectors where consumers picked up agentic tools earliest. The bots comparison-shop, scrape streaming catalogs, book flights. Training crawlers still made up 67.5% of AI-driven traffic, but their share dropped throughout the year as a newer category, AI scrapers, grew 597%. Media and streaming sites absorbed 41% of scraper traffic. E-commerce took 37%.

"AI-driven traffic is no longer experimental," Stu Solomon, HUMAN Security's CEO, told CNBC. "It's becoming embedded in core digital customer experiences."

And the agentic category barely existed twelve months ago. That 7,851% growth rate started from almost nothing in 2024. Small base, enormous trajectory. AI agents are not just browsing anymore. According to HUMAN Security's data, a growing number now access user accounts and push into checkout flows, performing transactions for the people who sent them.

The 1,000x multiplier

Prince, whose company serves roughly one-fifth of all websites globally, put the math in plain terms at SXSW. A person shopping for a digital camera might visit five websites. An AI agent performing the same task could visit 5,000.

"That's real traffic, and that's real load, which everyone is having to deal with and take into account," Prince said at SXSW.

Before generative AI, bot traffic sat around 20% of the internet. Googlebot still ingests more web content than all other AI crawlers combined, pulling 4.5% of all HTML request traffic versus 4.2% for every other AI bot put together. But OpenAI's GPTBot, which roughly quadrupled between May 2024 and May 2025, is eating into that lead. Its ChatGPT-User crawler, the one grabbing real-time answers for live chat sessions, grew 2,825%.

Prince expects bot traffic to pass human traffic by 2027. COVID-era traffic spiked hard and then leveled off after two weeks. This growth hasn't.

"We're seeing internet traffic grow and grow and grow," Prince said. "And we don't see anything that's going to slow it down or stop it."

By Cloudflare's reckoning, global internet traffic grew 19% last year alone. The curve shows zero sign of bending.

Publishers and the vanishing click

The shift hits content publishers hardest. TollBit, a firm that tracks AI bot activity on publisher sites, reported in February that by the fourth quarter of 2025, there was one AI bot visit for every 31 human visits. In the first quarter, the ratio had been one to 200. Human web traffic actually declined 5% between the third and fourth quarters.

What changed was not just volume but composition. Training scrapes, the kind that hoover up content to build foundation models, dropped 15% between the second and fourth quarters of 2025. The bots replacing them were retrieval-augmented generation crawlers, the systems that fetch real-time answers for ChatGPT, Gemini, and their competitors. RAG bot traffic grew 33% in the same period. AI search indexers jumped 59%.

The problem for publishers is simple. These bots take content but send almost nobody back. Clickthrough rates from AI applications to the sites they sourced fell from 0.8% in the second quarter of 2025 to 0.27% by the fourth quarter. Nearly a threefold collapse. Even publishers who signed licensing deals with AI companies fared poorly, watching their clickthrough rates drop to 1.33%, a 6.5x decrease from earlier in the year.

"AI traffic will continue to surge and replace direct human visitors to sites," Olivia Joslin, TollBit's chief operating officer, told The Register. "Ultimately, AI will become the primary reader of the Internet."

Thirty-seven percent of active AI users now start their searches in AI platforms rather than Google, according to marketing firm Eight Oh Two. Pew Research found that 62% of U.S. adults use AI in some form at least several times a week. You can feel the consequences in any publisher's analytics dashboard. The audience is not disappearing. It's being intermediated.

Prince, at SXSW, was blunt about the business model. "Bots don't click on ads."

He sees a possible lifeline for media companies that produce original, hard-to-replicate reporting, particularly local outlets. "If you don't have the Park Record, then you don't get that information," Prince said, referencing the small Utah newspaper. "We may make more off licensing our content to AI companies than we do off digital advertising." That math only works if licensing fees materialize at scale, and the TollBit clickthrough data suggests they have not.

The security question nobody wants to answer

Not every bot browsing the web is comparison-shopping for cameras. The Imperva 2025 Bad Bot Report, published last April by Thales, found that automated traffic surpassed human activity for the first time in a decade in 2024, hitting 51% of all web traffic. Bad bots hit 37% of all internet traffic in 2024. They were at 32% a year earlier. Credential stuffing, payment fraud, data scraping. Six straight years of growth, each worse.

HUMAN Security's own numbers look grimmer on the attack side. Four hundred thousand post-login account compromise attempts per customer on average in 2025. That's four times the 2024 count. Retail and e-commerce absorbed 440,000 unique threat profiles, four times the runner-up industry. Seven in ten scraping attacks hit retail.

Cloudflare's 2026 threat report adds another layer. Ninety-four percent of all login attempts on its network now come from bots. Sixty-three percent of those use passwords already stolen from other breaches. The company blocks 230 billion threats daily.

And the arms race between publishers and AI scrapers intensified in parallel. According to Cloudflare data, 13.26% of AI bot requests ignored robots.txt directives in mid-2025, up from 3.3% in the fourth quarter of 2024. Publishers responded with 336% more blocking measures, and 79% of top news sites now explicitly block AI training bots in their robots.txt files. The bots keep coming anyway.

Lindsay Kaye, HUMAN Security's vice president of threat intelligence, described the anxious middle ground organizations now occupy. "Unquestionably trusting novel technology like agentic AI could lead to risks such as compromised credentials, data misuse, and unintended consequences when shopping," she said. The challenge is that blocking AI agents entirely means losing the customers who send them.

Measuring something that doesn't want to be measured

None of these reports claim to capture the full picture. Filippo Menczer, a professor of informatics and computer science at Indiana University who studies AI bot behavior, cautioned that estimates depend heavily on methodology.

"You can try to estimate the amount of bot traffic by looking at the agent strings, but these are very noisy estimates," Menczer told CNBC. "They depend on what sample you get."

User-agent strings are self-reported labels from web crawlers, and HUMAN Security acknowledged in its report that "the reliability of that self-identification is a growing concern." TollBit's own analysts noted that many AI scrapers are now "indistinguishable from human visitors on sites," making their published ratios conservative at best.

HUMAN's dataset comes from its customer base, not the entire internet. Imperva's from 13 trillion blocked bot requests across thousands of domains. Cloudflare's from the roughly one-fifth of all websites that use its infrastructure. Each slice is enormous. None is complete. But the direction every dataset points is identical.

From "bot or not" to "trust or not"

Solomon, the HUMAN Security CEO, framed the question facing every company that operates online. "This notion of machine bad, human good just is not realistic," he said. "You have to live in a world where machines are acting on our behalf, and we have to establish a level of trust that's persistent over time."

Solomon sees that shift as structural. Binary detection, bot or human, gave way to something messier. Continuous trust evaluation, he calls it. Was the visitor legitimate? Did it behave consistently across the session? Those are the questions that matter now, not whether the request came from a person or a script.

Prince described companies adopting "three radically different strategies" for dealing with AI agents. Some block bots entirely. Some welcome them. Some have not decided. That split tells you where the industry stands. Nobody has cracked this yet.

One thing is clear: the physical internet needs work. Prince envisions millions of lightweight computing "sandboxes" spun up and torn down every second to service AI agents. Disposable server architecture, basically. It does not exist at the scale he imagines. The cables and data centers and cooling systems will need to handle traffic that keeps growing without any of the plateaus that followed every previous internet surge.

The internet's founding assumption, as Solomon framed it, was that a human sits on the other side of the screen. It has been thirty-four years since Berners-Lee put up the first website. That assumption is gone. What fills the vacuum is still being negotiated. Companies selling trust infrastructure want their cut. AI platforms want content. Publishers want their audiences back. And the humans who kicked off this whole shift by asking a chatbot instead of opening a browser tab? They mostly just want answers.

Google's Quiet Conquest: What Cloudflare's Data Actually Reveals About AI's Power Grab

Cloudflare's sixth annual Year in Review reported global internet traffic surged 19 percent in 2025. Most coverage focused on the top-line growth, the record DDoS attacks, the post-quantum encryption

The Implicator

ChatGPT's U.S. Mobile Market Share Falls Below 50% as Gemini and Grok Gain Ground

The chatbot market more than doubled in size over the past year, and ChatGPT lost its grip on it. Two separate analytics firms, tracking different platforms with different methods, landed on the same

The Implicator

The Web’s New Paywall: Cloudflare Makes AI Bots Ask (and Pay) for Access

Good Morning from San Francisco, Cloudflare just declared war on freeloading AI bots. The company that handles 20% of internet traffic now blocks AI crawlers by default and launched a marketplace whe

The Implicator

Marcus Schuler

San Francisco

Tech translator with German roots who fled to Silicon Valley chaos. Decodes startup noise from San Francisco. Launched implicator.ai to slice through AI's daily madness—crisp, clear, with Teutonic precision and sarcasm. E-Mail: [email protected]