Some of your website's visitors aren't people anymore, and they aren't the old-fashioned bots you already know about either. HUMAN Security's 2026 benchmark report found that traffic from AI agents and agentic browsers grew 7,851% year over year. That number sounds alarming. It's also easy to misread. Here's what an AI agent actually is, what the data says, and what you should do about it if you run a site.

What an AI agent is (and how it differs from a crawler)

For years, "AI traffic" mostly meant crawlers. These are bots that collect pages so a model can train on them or so a search index can stay current. They read your content and leave. An agent is different. Wikipedia describes an agentic browser as one that can take actions – such as navigating webpages or filling out forms – on behalf of the user.

Put simply, a crawler reads your page, while an agent uses it. Someone asks an assistant to compare three hotels, find a product in their size, or book a table. The agent opens your site, clicks around, and reports back, or finishes the task itself. HUMAN sorts AI-driven traffic into three groups: AI training crawlers, real-time AI scrapers, and agentic AI (AI agents and agentic browsers).

What the HUMAN Security numbers actually say

The report came out on March 26, 2026, and it covers 2025. A few figures give the headline its context:

  • Automation is growing faster than people are. Automated traffic grew 23.51% year over year in 2025, compared with 3.10% growth in human traffic.
  • Crawlers still make up most AI traffic. Training crawlers make up the majority at 67.5% of AI-driven traffic, but their share declined sharply throughout the year as traffic from AI scrapers grew 597%.
  • Agents started from almost nothing. HUMAN's CEO said about half of online traffic is now machine-based, with roughly 2% classified as agentic activity acting with authority on a person's behalf.
  • Browsers are growing faster than standalone bots. Within the category, agentic browsers grew faster than purpose-built agents, roughly doubling agent traffic by late Q4.
  • Shopping is where agents go most. Retail & E-commerce leads agentic traffic at 46.6%.

So the honest takeaway is this: agent traffic is still a small slice, but it's growing very fast from a tiny base. A 7,851% jump doesn't mean agents outnumber your human readers. It means a whole new category of visitor showed up in a single year, and it's heading for the pages where money changes hands.

Companies are deploying agents on their side too. KPMG's Q1 2026 AI Pulse Survey of US leaders at billion-dollar-plus organizations found 54 percent actively deploying AI agents. Publishers have started adjusting. Digiday reported this week that People Inc. is building custom APIs so its journalism is easier for AI agents to use, and the report cited the 7,851% growth figure as the reason it's urgent.

Why you can't easily tell agents from people

This is where it gets awkward for site owners. Older bots announced themselves with a name in the user-agent string. Agents often don't. A peer-reviewed poster at the October 2026 ACM Internet Measurement Conference notes that agents typically operate using real browsers and perform everyday tasks, making them difficult to detect.

The researchers tested seven agents against commercial bot detection and found that only one of the seven AI browsing agents are detected. They also found that AI browsing agents can have identical browser fingerprints with each other and with humans.

That affects your analytics in a practical way. Some "visitors" in your reports may be agents running errands for real people, and your dashboard probably can't separate them. It's the same measurement problem we covered in our breakdown of zero-click search. Value is moving to places your standard traffic reports don't show well.

There's a security side for the people using agents, too. The biggest risk is indirect prompt injection, where instructions hidden inside a webpage trick the agent into taking unintended actions. Firecrawl's 2026 guide reports that Anthropic found that unmitigated agents fall for about 24% of prompt injection attempts. If you let an agent shop or fill in forms for you, keep a human approval step before anything involving payment.

The identity fix: signed agents and Web Bot Auth

The industry's answer is to make agents prove who they are. Cloudflare's docs describe Web Bot Auth as an authentication method that leverages cryptographic signatures in HTTP messages to verify that a request comes from an automated bot. Cloudflare now uses it so websites and bot creators use Web Bot Auth to segment agents from verified bots, making it easier for customers to allow or disallow the many types of user and partner directed bots.

It's useful, but it's early. One August analysis pointed out that its working group has no adopted documents — only individual drafts — while Cloudflare, AWS WAF, Akamai, HUMAN and Vercel already verify its signatures in production. A signature also has a limit: a valid signature identifies the operator, not the person. The standard is still changing, so check Cloudflare's own documentation for the current details before you change any settings.

What to do about it this month

You don't need a whole agent strategy. You need a few sensible defaults:

  1. Look at your bot settings. If your site runs behind a CDN or hosting firewall, see how it treats AI crawlers versus agents. Blocking every unfamiliar bot can also block a customer's assistant that's trying to buy from you.
  2. Decide crawlers and agents separately. Opting out of model training and turning away an agent that's shopping for someone are different decisions. Treat them that way.
  3. Make your key facts easy to read without a person. Agents go for prices, availability, specs, and policies. Put those in plain HTML text, not buried in images or behind interactive widgets.
  4. Expect your analytics to be a little wrong. Odd sessions with fast, linear clicks and no scrolling may be agents. Note it, but don't panic over one month's numbers.
  5. If you use agents yourself, keep a human in the loop for logins, payments, and anything you can't undo.

FAQ

Is an AI agent the same thing as a chatbot?

No. A chatbot answers in a chat window. An agent takes actions: it opens pages, clicks, fills in forms, and sometimes finishes a task like booking or buying. Many chat assistants now include agent features, which is why the line keeps blurring.

Should I block AI agents from my website?

Usually not across the board. Most agents act for a real person, often someone who wants to buy something or find information. A better approach is to handle training crawlers, scrapers, and agents separately, and to allow verified agents where your CDN supports it.

Does the 7,851% figure mean bots outnumber humans on my site?

No. It measures how fast agent traffic grew, not how big it is. HUMAN's own leadership put agentic activity at roughly 2% of online traffic. It's a fast-growing slice, not most of your visitors.