Helpin
Dark mode
Support content sources

Support content sources

Support content sources are the specific materials the AI reads when answering. Understanding the types and how to keep them healthy is the difference between a deflecting AI and a confusing one.

See also: Linking knowledge sources for where to connect sources in Settings → Knowledge. This page focuses on choosing and maintaining them.

Source types

Source

What it is

Good for

Docs spaces

Helpin documents in connected spaces

Product how-tos, policies, internal runbooks (internal spaces)

Help center

Your published public articles

Customer-facing answers; widget Help tab

Website sources

Public pages crawled from URLs

Content that lives outside Helpin — existing docs sites, changelogs, feature pages

Choosing what to connect

Connect content that is:

  • Factual and current — the AI cites what you give it; stale content produces stale answers.

  • Customer-phrased — articles answering real questions beat internal jargon for deflection.

  • Scoped — don't connect everything "just in case"; irrelevant sources crowd out relevant retrieval.

Keep internal content (runbooks with internal links, security details) in internal spaces if you don't want it surfaced to customers via the widget Help tab — AI answers cite published sources.

Maintaining sources

  • Publish to update — republished docs refresh the index; drafts don't.

  • Retire dead content — unpublished or deleted articles drop out of citations.

  • Re-crawl website sources when the underlying pages change materially.

  • Regenerate the company context after major product changes (Knowledge → Regenerate with AI, then edit).

How retrieval uses sources

When a customer asks something:

  1. The question is matched against indexed content across all connected sources.

  2. Top passages from any source type are candidates.

  3. The AI answers from the best material and cites it.

There is no per-source priority switch — relevance decides. That's why source quality and overlap matter more than quantity.

Signs a source set needs work

  • AI answers citing the wrong article → overlapping or vague content; rewrite the canonical article.

  • Repeated handoffs on the same topic → missing article (check Coverage Gaps).

  • Answers framed with outdated product names → stale content or stale company context.

Was this article helpful?