Live on this site
Your next buyer sends an agent first.
An assistant reads your pages, compares you against three others and reports back — often days before any person visits. Most analytics records that as bot traffic and deletes it, then treats the human who follows as a stranger with no history.
This page shows what we do instead. It is not a mockup: the simulator below runs the same code that handles real traffic here.
Try it
Choose a visitor. Each runs through the real classifier, the real attribution resolver and the real handoff logic. Nothing is written to our database — the verdict is computed and discarded.
AI agents
The human who follows
Not welcome
Pick a visitor.
Each one runs the same classification, attribution and handoff code that handles live traffic on this site. Nothing is written to our database — the simulator computes a verdict and discards it.
How an agent is identified
A user-agent string is a claim, not an identity. Anyone can send ChatGPT-User. What makes it real is a forward-confirmed reverse-DNS lookup: the address is resolved to a hostname, the hostname must belong to the agent operator, and that hostname must resolve back to the same address.
Only verified reads are ever recorded. That is the whole reason a forged header cannot buy priority treatment later — there is nothing for it to earn.
- Verified agent — served fully, no consent banner, no cookie
- Claimed but unverified — served, not trusted, nothing recorded
- Automation — minimal response, no capture path
- Human — full page, consent banner, capture
Not every agent means the same thing
Treating all bots as one category throws away the signal. The distance between a crawler and a live fetch is the distance between “someone may read about you eventually” and “someone is asking about you right now”.
| Agent | What it means | Lag to a human | We keep the read for |
|---|---|---|---|
GPTBot crawl | Bulk crawl for a training corpus | weeks to months | 90 days |
OAI-SearchBot index | Building a search index | days to weeks | 30 days |
ChatGPT-User live | Fetching a page to answer someone mid-conversation | minutes | 6 hours |
Joining the agent to the person
When a verified agent reads a page and a human arrives on that same page from that agent’s product shortly after, those are not two visitors. They are one buying party, and everything the agent read belongs to that party’s journey.
The part nobody else does
An agent can enquire on its user’s behalf — and get something useful back
Reading a site is passive. An assistant shortlisting vendors needs to ask a question and get an answer it can repeat. So we expose one: describe what your user needs, and get matched services, the terms that matched, and an honest fit assessment — including when the answer is that we are not the right people.
| Endpoint | What it does |
|---|---|
GET /.well-known/agent.json | Discovery — what we offer and how to ask. |
GET /api/agent/services | The full catalogue: scope, outcomes, engagement model, and what we do not do. |
POST /api/agent/enquiry | Describe a need, receive matches with reasoning and a handoff link for your user. |
Try an enquiry
This calls the live endpoint. Note the pricing: indicative bands go to agents, and are deliberately not published on the service pages. Software scope is too variable for a headline figure to help a human reader — it anchors the conversation before anyone understands the work. An agent, though, cannot shortlist a vendor it cannot compare.
Send an enquiry
Pick an enquiry.
This calls the real endpoint. The reply below is exactly what an assistant receives — including commercials that are not published anywhere on this site for human readers. Where a standing rate exists it is quoted; where it does not, the reply says how the work is priced rather than inventing a range.
The handoff link is the point
The reply contains a URL to give your user. It carries an opaque token — no name, no email, nothing that identifies anyone. When they follow it, minutes or weeks later, on a different device, the research you already did joins their arrival deterministically.
No six-hour window. No guessing from a referrer. The join key travelled with the recommendation.
What we do with contact details an agent sends
We record them and hold them. An assistant saying someone is interested is not that person saying it, so the lead is not routed to anyone until a human turns up behind it — by following the handoff link or contacting us directly. Say so to your user; it is what actually happens.
What an agent gets here
The same facts as a person, in a form that does not require running JavaScript or guessing at prose. Both are generated from one content source, so an assistant cannot quote us saying something the site does not say.
- /llms.txt — the offer, scope and answers as plain text
- schema.org JSON-LD on every page
- /robots.txt — AI crawlers explicitly allowed, not merely tolerated
Blocking AI crawlers removes the top two stages of your own funnel. We do not recommend it, and we do not do it.
What we do not claim
That agent-preceded traffic converts better is a hypothesis. It is a good one — a live fetch means someone is asking about you now — but it has not been measured, and a flag that both predicts an outcome and changes the experience can never be evaluated afterwards.
So we record the signal, keep a control group, and weight it only once the lift is a number. If someone sells you an AI-traffic score without a control group, ask how they know.
Want this on your site?
This is the work we do — identification, machine representation, agent-aware capture, and reporting that treats the agent and the person as one party.