The chat knowledge base
Three ways content gets indexed, and how to check what the bot actually reads.
Chatbot & Live ChatEnterprise
Requires an Enterprise plan.
The bot only answers from content you've given it. Three sources feed the index:
| Source | Where it comes from | When to rely on it |
|---|---|---|
| Crawl | Fetched HTML from your site, started from the Knowledge tab. | Server-rendered sites — the fastest way to bootstrap. |
| Snapshot | Text indexed from real recorded page views. | JavaScript-rendered sites, where a crawler sees an empty shell but a real visitor's DOM has everything. |
| Manual docs | Documents you write or paste in yourself. | Policies, pricing rules, and anything not on your public site. |
Snapshot indexing is why chat works on SPAs where crawler-based tools fail: DeepSync already recorded the rendered DOM, so it doesn't need to render your app again.
Inspecting what was indexed
Open any page in the Knowledge tab to see the exact chunks stored for it. This is the only reliable way to catch boilerplate — nav menus, cookie banners, footers — that got indexed and is now diluting answers, or to spot a page that should never have been indexed at all.
If the bot gives a confidently wrong answer, look at the indexed chunks before rewriting the persona. Nine times out of ten it's answering correctly from stale or wrong source text, and the fix is a manual doc that overrides it.
Related
Still have a question?
Support is included on every plan, including Free. Tell us what you're trying to do and we'll point you at the right page — or fix the docs.
Contact support