Bots Now Account for More Web Traffic Than Humans
Bots Now Account for More Web Traffic Than Humans
By Aaron Rose · Tech Reader Magazine · July 3, 2026
The web has always had bots. From the earliest days of search engine crawlers methodically moving from page to page, automated traffic has been part of how the internet works. What is new is the ratio.
For most of the web's history, the balance was not in question. People visited pages. Bots indexed them. The gap was wide enough that publishers, advertisers, and platform builders could operate on a simple assumption: the entity on the other end of a request was a person.
That assumption is now being measured, and the measurements are drawing attention.
What the Numbers Show
Several major reports published in 2025 and 2026 have attempted to quantify the current bot-to-human ratio across web traffic. The figures vary depending on methodology and what slice of the internet each organization measures, but the direction is consistent.
Imperva's 2026 Bad Bot Report, drawing on 2025 traffic data across thousands of domains, found that bots accounted for 53 percent of measured web traffic — the second consecutive year automated traffic has outnumbered human activity. The year before, the figure was 51 percent.
53%Share of measured web traffic classified as automated in 2025, per Imperva's 2026 Bad Bot Report — the second consecutive year bots have outnumbered human activity. In 2024 the figure was 51 percent.
Cloudflare's measurement, drawn from its global network in June 2026, arrives at a different number: roughly 35 percent of web requests classified as automated, with humans accounting for 65 percent. Cloudflare measures a broader, more neutral cross-section of the internet. Security vendors like Imperva and HUMAN Security weight toward the application-layer surfaces they defend — login endpoints, checkout flows, APIs — where bot activity is more concentrated. Both figures can be accurate simultaneously. They are measuring different things.
What all three organizations agree on is the direction of travel.
The Composition Has Changed
Within the bot category, something has shifted in the past two years that is worth understanding separately from the raw percentage.
The traditional bot population was relatively stable: search engine crawlers, uptime monitors, security scanners, and a persistent layer of malicious bots. These have not gone away. But a new category has grown alongside them at a pace that analysts describe as without precedent.
AI-related traffic — crawlers harvesting content to train models, bots retrieving content to answer queries in real time, and agentic systems performing tasks autonomously on behalf of users — now represents a fast-growing share of automated traffic. HUMAN Security, which analyzed more than one quadrillion interactions across its platform, found AI-driven traffic growing eight times faster than human traffic across 2025. Agentic AI traffic specifically — bots acting on behalf of users rather than scraping for training data — grew approximately 7,851 percent year over year.
7,851%Year-over-year growth in agentic AI traffic in 2025, per HUMAN Security — bots acting on behalf of users, browsing, retrieving, and acting autonomously rather than scraping for training data.
Cloudflare's current data identifies Anthropic's crawlers as the second-largest bot operator by verified traffic volume, behind Google and ahead of Meta and OpenAI. That is a recent development, and it reflects the scale at which AI companies are now engaging with web content.
The Referral Question
One data point that has drawn particular attention from publishers is not the volume of bot traffic but what that traffic returns.
Cloudflare publishes crawl-to-referral ratios for major bot operators — the number of pages a crawler visits for every one visitor it sends back to the open web. Google's ratio sits at approximately 5:1, reflecting its longstanding role as a traffic source: for every five pages Googlebot crawls, it sends roughly one visitor back through search results.
The Ratio, Illustrated
Perplexity, which surfaces source citations in its answers: approximately 186 pages crawled per referral sent back.OpenAI: approximately 848:1.Anthropic's crawlers, as of mid-2026: approximately 4,580:1.Google: 5:1. These are not percentages. They are ratios — pages read versus visitors returned.
These ratios describe a structural feature of how AI assistants use web content. When a user asks a question and receives an answer inside a chat interface, the source pages that contributed to that answer are generally not visited. The content was read; the publisher received no visitor. This is not a flaw in any individual system's design. It is a consequence of how AI-assisted answers work.
A human shopping for a camera visits five websites. The AI agent doing the same task visits 5,000. The human generates five pageviews. The agent generates none.
The Attention Economy Before AI
It is worth noting that the web's advertising-based economic model was under pressure before AI crawlers became a significant part of the picture.
The attention economy — the framework in which publishers produce content and advertisers pay to reach the humans consuming it — had been losing structural integrity for years. Social platforms aggregated audience attention more efficiently than individual publishers could. Mobile browsing shifted time spent away from the open web. Search result pages increasingly answered queries without requiring a click. The human traffic that had sustained web publishing was already fragmenting across platforms and intermediaries long before AI entered the conversation.
Analysts who have tracked these trends suggest AI represents a new development in a longer arc, operating at a different scale and through different mechanisms — not an isolated event.
What It Means for Someone Publishing Content
If you are a content creator publishing to the web in 2026, some portion of your audience is not human in any traditional sense. How large that portion is depends on your site, your topic, your traffic sources, and which measurement methodology you find most credible. The estimates range from roughly a third to more than half of all requests.
Analytics that count pageviews are measuring something, but what they measure has become more complicated to interpret. A spike in traffic may reflect human interest, crawler activity, or both. The tools for distinguishing between them are improving but are not yet standard across publishing platforms.
For a human reader arriving at a page in 2026, the experience of the web is largely unchanged. Pages load, articles appear, videos play. The infrastructure behind that experience is increasingly engaged in a parallel set of transactions — between automated systems reading, indexing, retrieving, and responding to content — that the human reader does not observe.
The Bot-to-Bot Horizon
A separate question, newer and less settled, concerns what happens as agentic AI systems become more common — AI assistants that browse, retrieve, and act on the web on behalf of users rather than having users browse themselves.
Cloudflare CEO Matthew Prince noted at SXSW in March 2026 that prior to the generative AI era, bot traffic was approximately 20 percent of the web, driven primarily by Google's crawler. The figures being measured now represent a significant shift from that baseline, and agentic systems are identified in every major report as the fastest-growing category within automated traffic.
What the web looks like when a substantial portion of browsing is performed by AI systems acting on behalf of humans, retrieving content from servers that are themselves increasingly automated, is a question the data describes but does not answer. The infrastructure being built around it — agent authentication, machine-readable content formats, payment mechanisms for automated access — suggests the industry expects the question to become more pressing over time.
A Note on the Numbers
Any single figure in this area should be held lightly. The methodology behind bot traffic measurement varies significantly across vendors, and each organization's data reflects the slice of the internet it can observe. The Imperva figure and the Cloudflare figure are both published in good faith and reflect different measurement approaches applied to different samples.
What the data, taken together, supports is a shift in the composition of web traffic that is large enough to be visible across multiple independent measurement efforts, and fast enough that figures published in 2024 are already being revised upward in 2026 reports.
The web was built for human readers. Measuring how much of its traffic still is — and what follows from that — is now an active area of research, infrastructure development, and industry discussion.
Sources
Data referenced in this article draws on the 2026 Imperva Bad Bot Report (Thales), HUMAN Security's 2026 State of AI Traffic report, and Cloudflare Radar's published traffic data for June 2026.
Tech Reader Magazine
TechReaderMagazine.com