How to Make Good Training Data for BOKBOK

Keep Related Information Close Together

BOKBOK understands your content best when related details are close to each other in the same place. For example, it is better to keep the name, price, and link for one product together than to split them across different parts of a document.

Blue bag: price 200 TK
Link: https://domain.com

In short: for each item (product, plan, service), try to keep all important fields — name, price, link, short description — close together.

Use Simple .txt Files When You Can

Our system can read PDFs and docs, but in practice plain .txt files perform better for many use cases. They are easier to parse, have fewer layout problems, and let you control exactly how information is written.

• Use .txt files for product lists, pricing tables, FAQs, and short explanations.
• Keep formatting simple: headings, bullet points, and clear labels like “Price:” or “Link:”.
• Reserve PDFs/docs for long documents only when necessary.

The cleaner and simpler the input, the easier it is for BOKBOK to find the right answer quickly.

Less but Clearer Data Beats Complicated, Noisy Data

Adding more and more content does not always make your chatbot better. For BOKBOK, concise and focused data usually works better than huge, complicated documents.

• Remove old or duplicate information that no longer applies.
• Avoid mixing many different topics in one long file.
• Write in simple, direct sentences instead of long, marketing-heavy paragraphs.

When the AI sees a smaller set of clean, up-to-date data, it can answer more confidently and with fewer mistakes.

Group Similar Data in the Same File

Similar type of information should live together. This helps BOKBOK see patterns and answer consistently. Instead of putting everything in one giant file, split by topic.

Instead of including all data in the same file, use multiple files with related data. This often performs better than one big, mixed document.