Website pages

Add pages of your website to the brand's knowledge, one at a time, by sitemap, or with Smart import. We read them for you and keep re-reading them, so the AI answers from what your site says today.

Where these pages live

Open Brand → (your brand) → Knowledge, go to the Website tab in the lower half, and choose Add → Website page or sitemap. Web pages and sitemaps have their own table there, separate from your files, pasted text and Q&A pairs — those live on the Sources tab beside it. Each row carries a Web page or Sitemap badge in the Type column.

They get their own tab because they are the only sources that change on their own: we re-read them for you, they can fail, and the table has to tell you when each one was last read and when it will be read next.

These pages belong to the brand, and everything that answers for that brand searches them: chat, phone calls, the reply suggestions your agents see in the inbox, and AI social posts.

Single pages vs sitemaps

You can add content in two ways:

Single pages

Add individual web pages that contain important information. Perfect for:

  • Product documentation
  • FAQ pages
  • Policy pages
  • Important announcements

Sitemaps

Add an entire sitemap to read many pages at once. This is ideal for comprehensive websites, but requires careful planning.

Sitemap best practice: Only add sitemaps whose content is relevant to your AI. Avoid sitemaps that contain every page on your site — focus on pages with stable, useful information.

Smart import

If you want your whole website in, start here rather than adding pages one at a time or handing over a sitemap wholesale. Smart import reads your site's navigation or its sitemap, picks about twenty-five pages that actually help a support AI, and imports them in the background over a few minutes.

Running it again later updates a page it has already read instead of adding a duplicate.

Choosing content wisely

Not all web content is equally valuable for your AI. Some pages can actually hurt performance:

Good content

  • Static documentation pages
  • Knowledge base articles
  • Product specifications that don't change frequently
  • Company policies and procedures

Problematic content

  • Blog posts (often time-sensitive and not relevant to customer questions)
  • Product pages with dynamic pricing or inventory (the AI might quote something out of date)
  • News sections
  • Event calendars
  • User-generated content areas

The double-edged sword

Pages make content easy to add, but can introduce misleading information. Your AI will answer from whatever content you provide — including outdated information, seasonal content, or pages that aren't relevant to customer questions.

For very large websites (hundreds or thousands of pages), consider creating a secondary sitemap containing only the most important, stable content pages. This gives you precise control over what your AI reads.

Reading the Website table

Five columns tell you whether a page is working and when it was last looked at.

Status

One of Ready, Not indexed, Reading…, Warning or Stopped.

  • Ready — read and searchable. This is the one you want.
  • Not indexed — nothing was stored from it, so the AI cannot answer from it at all.
  • Reading… — we are working through it right now.
  • Warning — the last read had a problem, and we are still trying.
  • Stopped — it failed three reads in a row, so automatic re-reading was switched off for it. Your AI still answers from the last copy that worked.

You can filter the table by status, and if anything has stopped, a warning appears as soon as you open the Knowledge screen.

Issue

What went wrong on the last read, in plain words — Site blocked our reader, Page not found (404), Page took too long to answer, Page has no readable text, Sitemap could not be downloaded, Sitemap lists no pages. Hover it for how to fix that particular problem and for the exact technical error underneath.

Indexed

How many searchable pieces were stored from that page. This is how you tell a page that was genuinely read from one that produced nothing: a dash, or the Not indexed status, means the AI cannot answer from it.

For a sitemap, the line underneath also counts its pages — for example 27 pages · 3 failed. Open Pages found on the row to see exactly which addresses were read and which ones failed, each with its reason.

Last read

The last time we read the page successfully — the copy your AI is answering from right now. Two extra notes can appear under it:

  • Unchanged since … — a later read found exactly the same text, so nothing needed re-indexing. The page is being checked, it simply has not changed.
  • Last try … — failed — the page has been failing since that last good read.

Next read

When we will read it again by ourselves: In 4 days, Within 6 hours, Retrying within 6 h · attempt 2 of 3, Stopped — read again to resume, Paused by you, or Paused — needs a plan.

The Added column is hidden by default on this tab, since the two dates that matter here are Last read and Next read. Switch it back on with the Columns picker above the table.

Keeping pages up to date

Reading one page again yourself

  • Use the Read again now action on the row.
  • A page can only be read again once every 5 minutes.
  • Useful when you have just changed the page and want the AI to catch up immediately.

Reading everything again

Read all again, above the table, re-reads every page and sitemap of the brand. It runs in the background — rows switch to Reading… and the table updates itself as each one finishes, so you can carry on working. Anything read in the last five minutes is skipped, and a brand with large sitemaps can take a few minutes to work through.

Automatic re-reading

  • Pages are re-read for you roughly once a week, and the schedule is checked every few hours.
  • If the text has not changed since the last read, nothing is re-indexed — the page just shows Unchanged since ….
  • Automatic re-reading runs for accounts on a paid plan or a trial. On the free plan the Next read column says Paused — needs a plan: a page is read when you add it and whenever you ask for it to be read again.

Pausing one page

For a page that never changes, the row menu offers Pause automatic re-reading. That source is then skipped by the weekly schedule and its Next read reads Paused by you. Read again now still works on it, and Resume automatic re-reading puts it back on the schedule.

When a page keeps failing

A single failed read no longer takes a page out of the rotation. Instead:

  1. The read fails — the status becomes Warning and the Issue column says why.
  2. We try again on the next runs. Next read shows Retrying within 6 h · attempt 2 of 3.
  3. After three failures in a row the source is marked Stopped and we stop reading it automatically.

Your AI keeps answering from the last copy that worked the whole time. To bring a stopped page back, fix whatever the Issue column describes and use Read again now — if it succeeds, the normal weekly schedule resumes on its own.

You get told when this happens. When a source stops, or when pages inside a sitemap start failing, the account owner receives an email and an in-app notification explaining what happened, why, how to fix it, and what happens if it is left alone. It is one message per organisation covering everything affected, and it is only sent when something actually changes — a source that keeps failing is not announced over and over.

What language these pages are searched in

A website page carries no language of its own. It is searched in the brand's language, set at Brand → Brand settings → General. The language decides which dictionary is used to match a question against the text — it does not translate anything.

Check your pages from time to time — filter the table by Warning and Stopped, and remove or fix the pages that keep failing.