Tuning the chat experience

The first half of the document editor decides what a reader meets before they type. The second half decides what happens once they do: how long the answers are, which buttons sit in the toolbar, whether anyone can email you from the chat. Almost all of it is per-document, so two assistants in the same workspace can behave quite differently.

Behavior: how the assistant answers

Default chat tone sets the starting response style. Chatty gives full-length answers with headings, context, and the document's natural formatting, and is the default. Succinct gives brief, scannable replies, often bullet points, with only the essentials — better for quick lookups and phone screens. Readers can switch between the two mid-conversation regardless of what you pick, so this is a starting point rather than a rule.

Memory is off by default. Turned on, it uses a logged-in reader's own earlier conversations with this same document as background context. It has no effect for anonymous visitors and never crosses between readers or documents. It is a different thing from Prior chats under UI Options, which lets readers browse their own history themselves.

Disable chat attachments hides the attachment button for this document even when attachments are enabled at the owner level. When they are allowed, a reader can attach a PDF or Word file as temporary context for their own conversation — it is not added to the document's training.

The Custom prompt field takes up to 2,000 characters of free-text instruction, with a live counter — something like "Always respond in formal English. If a question is outside this document's scope, recommend the user consult their advisor." Your instructions layer on top of Docutrain's core rules, so grounding, citations and refusal behavior always win. On save, the prompt is screened by an automated safety reviewer ("Validating your prompt with our safety reviewer…"), and a rejected prompt comes back with its category and reason so you can edit and try again. Tone, language, form of address and formatting preferences are fine; instructions that override grounding, fabricate facts or direct abusive behavior are not. A reader's Succinct choice also still wins: ask for long detailed answers all you like, a visitor who switches to Succinct gets the short format.

What appears on the chat page

The UI Options tab controls the chat interface, across four sub-tabs.

Layout covers page chrome. Allow layout toggle lets desktop readers switch between the sidebar and classic layouts; it is off by default, and mobile always uses classic. Document selector adds a menu for browsing other documents from the same owner, also off. Landing page link shows a "Back to [Owner]" link at the top of the chat, on by default, and only where your plan includes a landing page.

Features is the long one — the tools readers get, grouped under headings. Under Content: Keywords cloud (on) shows clickable topic keywords; Downloads (on) exposes the files from the Attachments tab; Prior chats (on) lets signed-in readers browse, continue and delete their own past conversations with this document, and is never visible to anonymous visitors. Under Media, Image gallery and Video gallery are both off by default and stay disabled until you have actually configured gallery items on the Images and Videos tabs.

Under Sources: Citations (on) appends source citations to answers. This toggle can be disabled with an explanatory note on documents built from several source files or extended by added-content retraining, where page references would not be reliable. References shows a References chip opening your curated bibliography, and References list is that bibliography, one entry per line — URLs and DOIs become links automatically, with a counter and a live preview. Under Overview, Abstract puts the abstract button in the toolbar, and stays disabled until an abstract exists.

Several of these accept optional English label and French label overrides: rename Downloads to "Handouts", the image gallery to "Slides", References to "Sources". Blank means the default. What readers do with the resulting toolbar is covered in chat tools and the sidebar.

Engagement, and the contact form's verification step

The Engagement sub-tab covers interaction. Recent questions (off by default) shows a gallery of questions others have asked, and needs at least two before it displays anything; Country flags adds flags based on asker location and appears only when recent questions are on. Quiz puts a Quiz button beside the keywords, and stays disabled until questions have been generated on the Quiz tab.

Contact form lets readers message you from the document. Messages go to the addresses in Recipient email, entered as pills: type an address, press Enter or comma to add it, click the x to remove one. The first pill is labeled To and receives messages; later pills are CC. The sender is set as reply-to, so you can answer directly. Intro text sets what appears at the top of the form.

Every recipient address has to confirm it is willing to receive contact-form mail, and verification is remembered per organization, so an address verified once is ready on your other documents. Each pill shows its state: Your email for your own sign-in address (verified automatically), a check mark once verified, Send link if it has never been asked, Link sent with a clock right after, then Resend once the cooldown passes. Invalid addresses show as red pills. Links expire after seven days, and while anything is pending a small refresh button checks status immediately.

The consequence matters more than the mechanics: if the To address is missing or unverified, the Contact tool simply does not appear on the chat page. Readers never see a broken form. Unverified CC addresses do not block anything — they are quietly left off the message until they verify.

Character is the fourth sub-tab, and appears only when the chat assistant feature is enabled for your organization. Enable character is off by default. Character name takes up to 60 characters and falls back to your owner-level name if blank; if neither level has a name but the character is on, readers see the generic "Assistant". Avatar can be uploaded or generated from a text description, which is kept so you can regenerate. Setting one here fully replaces the owner's character for this document.

Where everything else lives

The Attachments tab manages the downloadable files themselves: upload, name and order them, and when the image gallery is active, individual attachments can also appear in the lightbox. A Download Activity card below the list tracks who downloaded what, when and from where, with CSV export.

The remaining tabs are covered elsewhere in this series, but it helps to know where they are:

One tab is super-admins only: Metadata, a read-only panel of timestamps, raw configuration, chunk counts and internal identifiers, kept for support and diagnostics. Owner admins never need it.

If you change one thing after training a document, make it the custom prompt — the cheapest way to make an assistant sound like yours without touching the content.