A small team ships a product with decent documentation and a support inbox that receives the same twelve questions on rotation.

The docs answer all twelve. Customers do not find them, because finding them means knowing which of four guides the answer lives in, and customers do not think in guides. They think in "I am trying to do this thing".

A grounded chat assistant on the docs site closes that gap without anybody writing a new word. Here is the order to do it in.

Public access first, or nothing else works

Every embed and share method delivers a visitor to the same chat, so the document's access level decides whether they can use it. For a public docs site that means Public.

Docutrain is blunt if you forget. The share tab shows a banner naming the current level and spelling out what anonymous visitors will actually hit: a login wall, a passcode prompt, or a refusal. Better than discovering it from a confused customer.

Public access is a paid feature. On plans without it the tile carries a lock.

Getting it onto the page

Three shapes, in two places.

Button and Inline are built in the document editor. A button embed opens the chat in a fullscreen slide-in panel, with a live preview while you set its color and label. Inline puts the chat directly into your layout, streamlined without the footer or document switcher, which suits a dedicated help page.

The floating bubble is a saved embed, built in the dashboard's Embeds tab, on Business and above. It is worth the extra step for one reason, which the tab states outright: each embed keeps its own settings, and editing one updates every page that uses it. No re-pasting snippets across eleven pages of a marketing site.

Changes save as a draft and reach your site only when you publish, propagating in about a minute. Statuses tell you where you stand, and copying the code is disabled while an embed is still a draft, because a draft's key resolves to nothing.

Two things to get right the first time. A bubble's target document cannot be changed later, so a bubble for a different document is a different embed. And deleting an embed destroys its key permanently: any snippet already on a website stops working, forever. See embedding on your website.

Making it look like it belongs

Set your logo and accent color in Owner Settings. The accent is used for buttons and highlights across every branded view, embedded widgets included, and matching it to your product's color is the difference between an assistant and a bolted-on third-party thing. See branding and landing pages.

Then decide what visitors are told before they type. Two settings do this and they are not the same thing.

The explainer is a short note saying answers come only from the trained material and that it will admit when it does not know. Off by default, shown once per session.

The disclaimer is a gate. Readers must accept before they can chat at all. For ordinary product documentation that is usually overkill, and it puts a click between a confused customer and their answer. For anything with contractual or safety weight it is the right tool.

One detail that catches support teams out: the explainer never appears inside an embed, because the host page provides its own framing. If you want customers on your docs site to know what they are talking to, write that sentence into your page copy or into the bubble's greeting. Do not assume the explainer is doing it.

Ten minutes that beat any amount of prompt tuning

Two tools turn other people's questions into signposts.

Recent questions shows a live list of what other readers asked. Duplicates collapse, questions the assistant could not answer are filtered out, and it stays hidden until at least two suitable ones exist. Clicking one asks it, and if the document has been retrained since, a fresh answer is generated rather than a stale one replayed.

Pinned questions is the curated version, and it is where your twelve rotation questions belong. Admins get a plus button on each recent question offering Pin Question and Delete Question.

Pinning the twelve is ten minutes of work and does more for first-contact resolution than any amount of prompt tuning. It converts your support inbox's actual contents into the first thing a visitor sees.

Escalation needs a route out, which is the contact form. Note that every recipient address has to confirm it is willing to receive that mail, and until the main address verifies, the Contact button does not appear at all. Nobody is handed a form whose messages would go nowhere, but do check that verification before launch rather than wondering why nobody is writing in.

What you learn once it is live

The direct signal is the thumbs-down, which opens a form where readers pick what went wrong. Those reasons arrive in the conversation viewer as tags, so a red dot in the list is your highest-value review queue.

On a support assistant, Missing references is the one to watch. It usually means the doc page exists but was never cited, which is a retrieval or a wording problem rather than a gap.

Some classifiers run automatically. Conversations are checked for profanity and junk, and anything caught is kept out of the public recent-questions list and out of the group-wide feed. It is fully automatic with no manual override, it deletes nothing, and it occasionally catches something legitimate like a bare number answering a follow-up. You can audit what it caught in the per-document view. See conversations and moderation.

For the aggregate picture, the Intelligence tab is where documentation gaps surface by name.

Keyword density counts the literal phrases customers type, which is frequently not the vocabulary your docs use. That mismatch alone is often the whole problem.

Once your organization has around 200 questions, the Content gaps list ranks topics by a gap score built from answers that found no source, got re-asked, or were thumbed down, weighted by how often people asked. That list is a writing backlog in priority order, handed to you.

Turn on Exclude owner & admin traffic before you read any of it, or you will be looking at your own testing.

One thing to sort out before launch

Each organization has a daily chat allowance across all its assistants and readers: 300 questions per rolling 24 hours by default, independent of plan tier.

When it runs out, readers see a banner saying the assistant has reached its daily limit and when capacity returns, and the input is disabled until then. It is a rolling window, not a midnight reset.

If a public docs widget is going to see real traffic, ask the Docutrain team to raise that allowance before launch rather than after your busiest morning.