Content best practices

Well-structured source material improves hybrid search, citation quality, and answer accuracy. These practices apply to uploads, URL imports, and crawls.

Write for retrieval, not just reading

Do	Avoid
Clear H1/H2 headings per topic	Long walls of text without breaks
One concept per section	Mixing unrelated policies in one page
Explicit definitions (“Refund policy: …”)	Vague pronouns without antecedents
Tables for comparisons	Critical facts only in images
Consistent product and plan names	Alternate spellings for the same term

Chatevo chunks on structure; headings become strong retrieval anchors in Qdrant.

Match how visitors ask questions

Add natural-language FAQ entries even in formal docs:

Q: How do I reset my password? A: Go to Settings → Security → Reset password…

The Query Strategist (Standard+ plans) rewrites vague queries — but concrete source phrases still help keyword and semantic retrievers align.

File and crawl hygiene

PDF/DOCX — use real text, not scanned images without OCR
Crawl — index /help/ and /docs/, skip checkout and account pages — Website crawl
Duplicates — remove outdated PDFs when publishing new versions; enable Auto-retrain for live URLs on Standard+
Classification — label sensitive sections confidential or restricted before attaching to public assistants

Tune after publishing

Upload or import content — Upload documents
Attach to assistant and run Test chat with 10–20 real visitor questions
Note misses and false positives
Adjust source docs or search settings (similarity_threshold, max_results)
Re-test until citations land on the right sections

Master prompt alignment

Tell the assistant how to behave when retrieval is thin:

Answer only from the knowledge base. Cite the document title. If no relevant passage is found, say you do not know and offer to connect the visitor with the team.

See Master prompt.