Try it, not a tour

Point it at your website. Watch it answer.

Not a recorded walkthrough of somebody else’s shop. Give us your address and the real agent reads your pages, then answers questions about your product with the source attached. No account, no card, no form first.

Takes about twenty seconds. We read up to eight pages and respect your robots.txt.

  1. 1

    Paste your address

    No account, no card, and nothing to install. Just the domain.

  2. 2

    It reads your pages

    Up to eight of them, live, honouring your robots.txt — you watch it happen.

  3. 3

    Ask it in the widget

    The real one, on a stand-in of your site, answering from what it read.

Or watch the film

Three minutes of it actually working

A real shop, a real fortnight: the widget answering an order question from the shop’s own pages with the source attached, the agent stopping itself when a case needs a person, one inbox with the reply already drafted, and the repeated questions turning into the next help centre article. No call, no signup.

Transcript

The full narration, if you would rather read it than watch. Each heading jumps the video to that point.

Someone lands on your site at eleven at night with a question you have answered four hundred times. Nobody is awake. They leave.

That is the whole problem. Not that the answer is hard — that nobody is there to give it.

Chattering is a support agent you embed on your site. It reads your own pages and help centre, answers in your words, and hands the hard ones to a person.

This is a real workspace. Larkspur Home, a small bedding brand, on an ordinary Tuesday.

Fifty conversations in a fortnight here, and the customer got an answer in about a minute every time — most of them without anyone at the company ever knowing it happened.

Start with the part your customers actually see.

The widget sits in the corner of your site. Someone asks where their order is, and the answer comes back in a couple of seconds, from your own tracking page.

No ticket, no form, no waiting until Monday. And it cites the page it took the answer from, so you can check its work.

You are not writing any of this. It is reading what you have already published and turning it into an answer.

Second, and this is the part most of these things get wrong. Knowing when to stop.

A duvet arrives with a mark on the seam. That is a fault, a replacement and a photo — not something to guess at.

So it does not guess. It apologises, it says a person will pick this up today, and it puts the conversation in front of you with everything it already knows attached.

You set that line yourself, in plain English. Refunds over a certain size, anything about damage, anything it cannot verify — those come to you.

Third, everything lands in one inbox.

Widget chats, emails, the lot — in one list, with the ones that need a person marked and the routine ones already closed.

Fourteen of these are still open, because the agent decided each of them needed a person. The other thirty-six closed themselves.

Open one, and you are looking at the whole conversation plus a reply the agent has already written from the same sources it answers with.

Send it, change it, or take the conversation over completely and it steps back.

And last, the part that tells you whether any of it is working.

Every conversation is scored: what got answered, what got handed over, how long it took, and what people kept asking about.

The questions that come up over and over become the next help centre article — which the agent then answers from. It gets better because you publish, not because you retrain anything.

One agent on your site, answering from what you have already written, and stopping where you tell it to.

Put it on your own site

One script tag. It reads what you have already published and starts answering.

Start free
AI Support Demo — See Chattering in 3 Minutes | Chattering.ai