Malicious bot detection
Detect AI agents. Stop the bad ones, not all of them.
AI agents now browse, sign up and buy for the people who sent them. Some bring you customers. Some farm your free tier. Detecting that a visitor is an agent is the easy part. The useful answers are which company runs it, whether it can prove that, and whether the browser behind it has already signed up under other names.
Why agent traffic is different
Agent traffic is growing fast: 7,851% growth in AI agent traffic in a year, according to HUMAN, 2026 State of AI Traffic.
Agents run in real browsers, from real cloud and residential IPs, on behalf of real people. Bot scores built to catch scrapers either let them all through or block the good ones with the bad.
Signed, declared and unknown
Heedeo verifies signatures and names the operator. A declared agent with no signature is a reason on its own, because it is a claim nothing backs. A headless browser is recorded, never a reason: plenty of honest tools are automated.
- Signed: the agent signs its requests with Web Bot Auth, so you can verify which operator sent them, for example chatgpt.com.
- Declared: the user agent says it's an agent, but nothing proves it. Anyone can type "ChatGPT" into a header.
- Unknown: nothing says agent, though an automated or headless browser may still show.
Who built it is not who it's for
Knowing a request came from a well-known agent tells you who built the agent, not who sent it. The same agent serves your best customer and someone farming your trials. So Heedeo looks past the agent: at the browser it runs in, how many identities that browser has shown you, the email and the network, and whether one operator suddenly sends a crowd.
What Heedeo gives you on every signup
- Verifies agents that sign with Web Bot Auth and names their operator. An unsigned claim to be an agent, or a sudden crowd from one operator, is a reason. Being an agent never is.
- Recognises each browser that reaches a signup or other action, from a script you serve on your own domain. A browser that clears its cookies is usually still recognised by its characteristics.
- Counts the distinct emails and accounts each browser has shown you. Crossing your threshold (3 by default) is a strong reason on its own.
- Adds the weights of every reason into a score, and flags the event when it reaches your threshold (100 by default). Every reason comes with one line of plain text.
- Your rules decide what each score gets, route by route: allow, restrict, challenge with a passkey or email link, or block. Real customers never see a puzzle.
Questions
- How can I tell if a visitor is an AI agent?
- Agents that follow the Web Bot Auth standard sign every request, so you can verify who sent it. Others can be recognised by user agent, which proves nothing on its own, or by signs of browser automation. Heedeo checks all three and reports them separately.
- Should I block AI agents?
- Not all of them. Agents now sign up and buy for real customers. Judge the signup by its behaviour: how many identities the browser behind it has used, the email and the network. Let verified agents with a clean record through.
- Which agents sign their requests today?
- ChatGPT agent, Goose, Browserbase and Anchor Browser were among the first to sign with Web Bot Auth, listed in Cloudflare's signed agents directory.
- Is this the same as AI crawler blocking?
- No. Crawler blocking (robots.txt, firewall rules) is about who may read your content. Heedeo is about who signs up and what they take: trials, credits and accounts.
Take heed. Let's set you up.
Heedeo is ready. Book a 30 minute demo, and we set it up for your site with you.