Many teams stop after connecting a website and then wonder why the chatbot is vague on SKUs, tariffs, or procedures. Those details usually sit in PDFs, FAQ exports, and help-center articles — exactly the files Oxidesk is designed to index.
Pick sources that already answer customers
Good training files are current, customer-facing, and specific:
- Product data sheets and installation notices
- Plan comparison tables and billing FAQs
- Return, warranty, and SLA documents
- Help-center articles that already rank in your search box
Skip meeting notes, old release dumps, and three-year-old PDFs unless you have reviewed them. The assistant will treat them as true.
Keep one canonical document per topic
If shipping is explained in four overlapping PDFs, the chatbot has to guess which one wins. Consolidate or mark a single source of truth. Citations then point customers to the page you actually want them to read.
Retrain when the file changes
A price list from last quarter is worse than no file at all. When you publish a new catalog or help article, replace the source and re-index. Oxidesk’s knowledge-gap report shows questions that still miss a source so you know what to upload next.
Upload files from your workspace at oxideskai.com after you have connected the public site. Website first, files second, weekly gap review after that.
Super Admin
Head of Customer Education at Oxidesk