CAPTCHA on your form no longer stops AI agents. What it means for a small business contact form
Proof of Human, a maker of bot detection, tested five bot detection systems on a five-question survey. Across all adversaries, Google reCAPTCHA v3 identified 69.4% of automated sessions, hCaptcha 44.3%, Cloudflare Turnstile 42.1% and Cloudflare Bot Management 9.1%. Turnstile and hCaptcha stopped nearly all classic scripted browsers but few AI-agent sessions. The figures come from a competing vendor, so take them with caution. The message still stands: the form on your website finds it harder and harder to tell a person from an AI agent.
What happened
Five “adversaries” each took the survey, with multiple-choice and open-ended questions, 20 times: a standard automated Chromium browser, CloakBrowser, which hides automation signals, and three AI agents that “see” the page and decide what to click (Browser Use, Perplexity Comet and Grok). The systems tested were Proof of Human, Google reCAPTCHA v3, hCaptcha, Cloudflare Turnstile and Cloudflare Bot Management.
The clearest split was between scripts and AI agents. Cloudflare Turnstile detected none of the three AI agents and did not even issue challenges during their sessions. hCaptcha served an image task, but the vision-capable agents mostly solved it. reCAPTCHA v3 caught most agent sessions but none from CloakBrowser. Proof of Human credits its own system with 99%.
The authors list their own limitations: they tested the product tiers available to them rather than enterprise versions, some systems crashed, so sample sizes differ, and they did not measure how many real people the systems wrongly flag as bots.
What it means for you
- CAPTCHA is one hurdle, not the whole defence. If your contact form, booking or survey relies on CAPTCHA alone, assume an AI agent can get through. Combine it with a limit on submissions, email confirmation and a manual look at suspicious enquiries.
- Decide what an agent may do on your site. Some AI agents are sent by your real customers to ask about a price or a slot for them. Decide in advance what an agent may do, for example send an enquiry, but not complete an order or booking without a human confirming.
- Watch what comes in through the form. Look for sudden spikes in enquiries, near-identical messages and odd addresses. If they appear, tighten the form rules first. Don’t switch protection tools based on a single vendor test.
2-Point Expert Read
Not sure how exposed your website is, or what AI assistants and search engines can see on it? Send me your link. Within 24 hours you get one finding about your website, one about your visibility on Google and in AI search, and a recommended next step. For small businesses and sole traders, no obligation.
Weekly AI Digest for Small Businesses
Once a week: AI news and what it means for your business.
Share this story
Sources
- Proof of Human Research: Benchmarking Bot Detection Systems Against Modern AI Agents in 2026 (in English)
The test was published by Proof of Human, which makes a competing product and credits its own system with 99%. How many real people the systems wrongly flag as bots was not measured.