How to train a chatbot on your own content (without writing a single FAQ)

The phrase "training a chatbot" conjures images of spreadsheets full of question-answer pairs, AI researchers, weeks of work, and a hefty invoice. For Promptly, training means pointing it at the content you already have and letting it do the indexing. No FAQs to write. No question-answer pairs to curate. No prompt engineering.

Here is how each knowledge source type works, what it is best for, and how the freshness problem gets solved automatically.

Source type 1: Web crawl

Give Promptly a URL — your help center, your documentation site, your main product site — and it will crawl it the way a search engine crawler would. It follows internal links up to five levels deep, extracts the meaningful text content from each page, strips out navigation and cookie banners, and indexes it all.

This is the fastest way to get a chatbot up and running. A help center with 200 articles is indexed in under five minutes. After that, the crawler runs on a schedule you configure — daily, weekly, or triggered via webhook when you publish a new article. Your bot always knows about the thing you updated yesterday.

Source type 2: Sitemap import

Most websites have a sitemap.xml — a structured list of every URL on the site. If yours does, you can drop the sitemap URL into Promptly and it will bulk-index every page on the list in one operation. This is faster than a crawl for large sites because there is no link discovery phase — every URL is explicitly listed.

Sitemap imports are ideal for migrations. If you are rebuilding your help center, adding a new product line, or onboarding a site that already has a year of content, the sitemap is the quickest path to coverage.

Source type 3: PDF and DOCX upload

Some of your best content lives in files, not on web pages. Product manuals. Onboarding guides. Internal process documents that explain how things actually work. Promptly can index PDFs and Word documents directly — it extracts text and tables, handles multi-column layouts reasonably well, and treats the content identically to web-crawled content.

• Useful for: technical manuals, compliance documents, product spec sheets, training materials.
• Limit: 50 pages per file on paid plans (beyond that, consider splitting the document or hosting it as a web page instead).
• Re-indexing: upload a new version of the file when the content changes — the old version is replaced.

Source type 4: Manual knowledge entries

Sometimes you want to add knowledge that does not exist anywhere else yet — a new policy, an answer to a question that keeps coming up, context about your pricing that is not published. Manual entries let you type directly into a form: title, content, category. Done. That knowledge is available to the bot immediately.

Manual entries are also useful for override scenarios. If your website says one thing but your actual current policy is slightly different, a manual entry with the correct information wins. The bot uses the most relevant source for any given question — it does not blindly return the first result it finds.

The freshness problem — and how auto re-crawl solves it

The most common complaint about chatbots is that they become stale. "Our chatbot told a customer our old return policy" is a real thing that happens when knowledge bases are set-and-forget.

Promptly solves this with scheduled re-crawls. For each web-based source, you configure a re-crawl interval. Daily is the default for most teams; weekly works fine if your content rarely changes. When the re-crawl runs, it detects which pages changed, re-indexes only those pages, and leaves unchanged content alone. The process typically takes a few minutes even for large sites.

Citation-backed answers: why they matter

Every answer the bot gives includes a "Sources" link that points to the exact page or document it drew the answer from. This is not cosmetic — it serves two distinct functions.

For customers: it allows them to verify the answer and read more context if they want. Customers who follow citations are more confident in the answer and more likely to trust your brand.

For you: the unanswered questions report in Promptly analytics shows you every question the bot could not find a good source for. That list is your content roadmap. The gaps tell you exactly what to write next.