Knowledge Base FAQ
This FAQ explains typical problems with the knowledge base, PDFs and website content. The short rule: a Zeptix bot answers best when the sources are short, text-based, unambiguous and thematically separated.
Why does my bot say "I don't know"?
The bot says "I don't know" when no matching sources are found or when the sources found cannot reliably answer the question. This is better than hallucination. Check the file structure, key terms and test questions.
Can Zeptix read image PDFs?
No, pure image PDFs without a text layer do not provide usable text. Run OCR first, for example ocrmypdf, Adobe Acrobat or another OCR tool, and then upload the text-based file.
Why are short sections important?
Short sections with clear headings can be matched to user questions more effectively. A paragraph should ideally explain one topic completely, instead of touching on three topics halfway.
Why do synonyms help?
Customers often use different words than your team. If your source says "Subscription" but visitors write "plan" or "membership", synonym sections in your knowledge base help.
Why should I repeat key terms?
Search benefits from unambiguous terms. If your document says "cost" once and later only "it" or "that", context is missing. Repeat product names, plan names and technical terms in each section.
How large can a PDF be?
The article library refers to 50 MB as a practical PDF limit. Large PDFs are nevertheless often worse than several small files, because topics should be cleanly separated.