Agent detection and identity

Agents are already on your site. Not all of them say so.

Agentic browsers like Instinct, Muse, Manus and Hark already read your pages, compare your prices and fill in your forms for the people who use them. Some sign every request. Some pass as an ordinary browser. Some send nothing at all. Agent Tracking recognises each one, names it even when it stays quiet, and records what it did.

See how we tell them apart ↓

Why it matters

Your analytics was built to count people.

Agents that never run your scripts never reach your dashboards. The ones that do are counted as people, which inflates sessions and drags conversion down. Neither version tells you which agent came, who it was working for, or what it was trying to do. That is the first question to answer before you design anything for agents.

Without 42A

  • Agent visits are filtered out as bot noise, or counted as people
  • A shopping agent and a scraper look the same
  • A demo request sent by an agent reads like one from a person

With Agent Tracking

  • Every agent visit named, with a confidence level and the evidence
  • Agents acting for a person kept apart from training crawlers
  • Leads, carts and bookings tagged with the agent that made them
  • You decide what is tracked and what 42A gets to see

How we tell them apart

Some agents announce themselves. Some work hard not to.

Detection is a ladder, from proof down to suspicion. Every visit is checked against each rung until one holds, and that rung is what we report. A hunch is never rounded up to a fact.

Rung What we read What it tells you Confidence
Signed A Web Bot Auth signature on the request, the standard agents like Hark and Muse now send. The request really came from that agent. Verified
Declared A known agent name in the user agent, like Manus or HarkBot. What the agent says it is. Anyone can copy a name, so it ranks below a signature. High
Tagged A name carried in the link or set by the agent’s owner. Who sent it, for agents that send nothing else. High
Fingerprinted The network it runs from and the exact browser build, matched to agents we have identified before. Which agent it is, by name, even when it never announces itself. Medium
Suspected Automation traces: a cloud network, a headless browser, the webdriver flag. Something automated is here, not yet who. People on a VPN can land here too, and the report says so. Low

No one catches every agent. One running a real browser on a home connection can look exactly like a person. When we cannot tell, the report says so, instead of guessing in your favour.

How it works

Recognise. Record. Decide.

One script or one middleware sees every visit. Agents are named, what they do is logged, and your rules decide what happens next.

Step 1 / 3 Recognise every visit
  1. Signature-Agent: hark.com · verified
  2. User-Agent: Manus · declared
  3. Cloud network + headless Chrome · suspected
Step 2 / 3 Record what agents do
  1. Muse · /pricing → /plans/scale
  2. Instinct, fingerprinted · sent the demo form
  3. Manus · called search_products on the page
Step 3 / 3 Decide what they may do
  1. Demo form · agent → allow, tag the lead
  2. Demo form · crawler → refuse
  3. Admin pages · any agent → refuse, with a reason

What changes

Know who is on the other side.

What you can see once every agent has a name.

A name for every agent

Signed, declared or fingerprinted, each agent gets an identity you can follow across visits, including the ones that never introduce themselves.

Agent traffic, counted apart

Which agents come, what they read and where they stop, kept separate from people so neither number distorts the other.

Every action attributed

Leads, carts, bookings and tool calls carry the agent that made them, so sales knows a request came from an assistant, and which one.

Where it fits

You can’t design a journey for a visitor you can’t see.

Agent Tracking stands on its own. It is also the first layer of 42A: once you know which agents arrive and what they try to do, you decide where to take them.

FAQ / Agent Tracking questions

What teams ask about Agent Tracking.

Straight answers about what we can see, what we can’t, and what we keep.

01

What is 42A Agent Tracking?

It recognises AI agents visiting your website, names each one with a confidence level and the evidence behind it, and records what it did. It works on its own, and it is also the first layer of 42A’s agent journeys.

02

Can it detect agents that try to hide?

Many of them. An agent that stops announcing itself still leaves traces: the network it runs from, the browser build it uses, automation flags, and a fingerprint we may have matched before. Once matched, it gets its name back, Instinct or Muse rather than “Chrome on a Mac”. Some will still pass as people. When we cannot be sure, the visit is labelled suspected rather than guessed.

03

Who decides what gets tracked?

You do, not us. Your team chooses what is tracked and what 42A gets to see: which pages, which events, which details. What you leave out is not sent to us. The same goes for the rules on what each agent may do.

04

How do we install it?

With one script on your pages, or a middleware in front of your site. The middleware also sees agents that never run JavaScript.

05

How is this different from bot protection?

Bot protection decides whether to block. Agent Tracking tells you which agent it is and what it wanted, so you can welcome the agents working for your customers and refuse the rest.

06

How does it connect to the rest of 42A?

Once you know which agents arrive and what they try to do, the other 42A solutions decide where to take them: a direct way in through Personal Agent Integrations, a page that adapts through AI-Powered Website, and live control through Brand Gateway.

Agent Tracking / 42A

Find out who is already visiting.

Turn it on for your site and see which agents came, what they did, and which ones never said who they were.

Explore integrations →