The agent answers from the sources you give it. Knowledge quality is the single biggest driver of reply quality, so invest here before tuning anything else.
Two knowledge base types
Heymarket supports two source types, both managed under Settings > AI > Knowledge Bases. You need at least one PDF or one website source before the agent has content to use.
PDF Knowledge Base. Upload PDFs the agent can reference. Useful for SOPs, onboarding guides, FAQ documents, and internal references that do not live on the public web. Limits: 50 PDFs per team, 50 MB per file.
Website Knowledge Base. Add up to eight URLs (with optional linked subpages) and Heymarket scrapes them on a schedule. Useful for help centers, pricing pages, and product documentation that change over time. When adding a website source, choose whether Heymarket should include linked subpages, then review the scraped pages before launch.
What to add
Start with the sources customers already use:
- Public help center articles
- Pricing and plan pages
- Product feature pages
- Onboarding and getting started guides
- Sales collateral and one-pagers
- Internal SOPs that describe how your team handles common cases
- For private content behind a login, export the material as a PDF instead of using a website source, unless your Heymarket setup explicitly supports that source.
If a teammate would point a customer at it, the agent should have it.
How to structure content for AI retrieval
The agent retrieves passages, not whole documents. Structure helps:
- One topic per article. Avoid grab-bag pages that mix unrelated information.
- Clear headings. Use H2s and H3s that name what the section answers.
- Plain language. Write the way a customer would describe the problem.
- No internal contradictions. Two articles that disagree will produce inconsistent replies. Pick one source of truth. For example, do not keep one pricing page that says free setup and another PDF that says setup fees apply.
- Customer-facing voice. Strip jargon, internal codenames, and engineering shorthand.
Set the rescrape cadence
Website KBs have a configurable rescrape period:
- Daily for high-velocity content (product pages with frequent updates)
- Weekly for most help centers
- Monthly for stable evergreen content
Use the Sync button to force an immediate refresh after a major content update. If a source is still syncing, wait before testing. If it fails, fix the source URL/file and sync again.
Content hygiene checklist
Before going live:
- Every source is current (no last-year pricing, no deprecated features)
- Duplicates are removed or merged
- Internal-only language has been rewritten for customers
- Conflicts between sources are resolved
- Sync status reads "Complete" for every source
Next steps
Continue to Article 4: Configure Agent Behavior.