The hosted page says the chatbot is not ready

What it is

Your hosted page has loaded, and the chatbot is politely explaining that it is not ready to answer questions yet.

That is untrained mode, and it is a specific, meaningful state — not a broken page. It means the chatbot is Active (so its URL resolves and visitors are let in) but has never completed a training run (so there is nothing for it to answer from).

Two states, two different symptoms, and telling them apart is the whole point of this page:

The chatbot is	Visitors get
Draft or paused	A 404
Active but never trained	The page loads, with not-ready copy
Active and trained	Normal answers

If you are seeing a 404 rather than a chat shell, you are in the wrong article — see The hosted page is 404.

When you would use it

• You shared the link and someone reported that the chatbot "does not know anything".
• You expected a 404 and got a working page with unhelpful replies.
• You uploaded knowledge and the page has not caught up.

Where to find it

Open Dashboard → Chatbots → your chatbot → Knowledge, and check Analytics for training readiness.

Why the page loads at all

It would be simpler to 404 an untrained chatbot. It would also be worse.

An active chatbot is one you have deliberately switched on, which usually means you are mid-setup and about to add knowledge. A visitor arriving in that window gets a branded page that says something honest, rather than a dead end that looks like a broken link. Nobody emails you to say your website is broken; they just leave.

The 404 is reserved for the state where you have explicitly said "this is not live" — draft or paused.

Steps

1. Open the /agent/{handle} URL. If it 404s, this is not your problem — activate the chatbot first.
2. Open Dashboard → Chatbots → your chatbot → Analytics and read the health card. Active plus never-trained is exactly the state described here. See Chatbot readiness.
3. Open Knowledge. Add something: pasted text, a file, a crawl, or a cloud import. See Knowledge overview.
4. Start small. One page of clean text trains in moments and proves the loop end to end.
5. Wait until the row reads trained. Pending and training mean it is not there yet. See Dataset status and training.
6. Open Test Chatbot as the owner and ask a question whose answer is only in what you just added.
7. Reload the hosted page in a private browsing window. Untrained mode should be gone.
8. Do not pause the chatbot to "fix" this. Pausing replaces a working page that explains itself with a 404, which is strictly worse.

What you will see

The public page loads a proper chat shell with your chatbot's name, avatar, and colours, and its replies are setup or onboarding copy rather than answers from your material.

On Analytics, the health card shows Active alongside a never-trained state. On Knowledge, sources are listed as pending, not started, or absent altogether.

The most common causes

Sources were added with training switched off. The Add drawer has a "save without training" route, and content saved that way sits at not-trained until you train it. Look for rows reading Not trained and use the train action, or the bulk train action in the banner.

Training is still running. A large PDF or a whole-site crawl takes minutes. The badge tells you.

Every source failed. A batch of scans with no readable text, or a crawl that collected nothing, leaves you with rows but no usable knowledge. Open a failed row for the reason — see Scanned documents and images if they were scans.

The knowledge went on a different chatbot. Easily done with two similar chatbots open. Check you are looking at the one whose URL you shared.

Limits and plan notes

• Untrained mode applies only to an Active chatbot. Draft and paused chatbots 404 instead.
• A chatbot that has trained before keeps answering while you add or edit knowledge. Updating never drops it back into untrained mode — the previous version stays live.
• Conversations in untrained mode still consume message credits, because the chatbot is still responding. Do not leave an untrained page shared and forget about it.
• The website widget behaves the same way on the knowledge side: it loads and cannot answer. It differs on the activation side, where it is blocked rather than 404ing. See Hosted page vs website widget.
• Removing every source from a trained chatbot returns it to this state.

Common problems

I added files and the page still says it is not ready.

Training may still be running, or the sources may have failed. Open Knowledge and read the badges rather than guessing from the public page.

Visitors see a 404, not untrained mode.

The chatbot is not Active. Untrained mode only exists on an active chatbot.

Test Chatbot answers but the public page does not.

Check you are testing the same chatbot, and reload the public page in a private window — your browser may be showing a cached earlier load. If it persists, look for an access gate you set and forgot.

It was working and has gone back to untrained.

You removed every source. Add knowledge again.

Should I pause the chatbot until knowledge is ready?

Only if you would rather visitors got a 404. If the link is already shared, leaving it active with the honest not-ready message is usually kinder.

How long until it clears?

As soon as one source finishes training. A short pasted page is quick; a large crawl is not.