Adding data sources to your chatbot
To get the most out of your AI chatbot, you need to make sure it has all the necessary context for your use case. Granularity and clarity of data sources that you add to your chatbot heavily impacts its response quality and accuracy.
In this guide, we will show step-by-step how to add different types of data sources to your Chatbit AI chatbot. While adding any source type is optional, we recommend adding at least your key website pages and the questions customers ask most often.
Where to add sources
Open your chatbot and go to Settings → Sources. The “Add training data” box accepts a website URL, files or plain text. Below it, the “Training data” list shows every source you have added, whether it is ready, and how many characters it uses against your plan’s limit.
After adding sources, click “Test chatbot” to ask a few questions in the Playground and check the answers.
Source types
You can add website pages, upload files and paste plain text. It’s important to understand that ultimately all sources that you add end up converted to plain text. Images and graphics are not read.
If your source files or website have a complex structure, you may want to cherry-pick the important information for your chatbot and add it as Text.
Website pages
Paste a page URL into the “Add training data” box and click “Add training data”. You can paste several URLs, one per line, and each becomes its own source. A bare domain such as example.com works too — Chatbit adds https:// and fetches the page.
Each URL is fetched, cleaned up and stored as text. The page address is kept with the source, so when the chatbot uses that page in an answer it can link the visitor to it.
When a page changes, click the refresh icon next to the source to fetch it again.
URL Mapper for whole websites
To add many pages at once, use the URL Mapper under Settings, or click “Map” next to any website source in your Training data. Enter a domain and click “Map website URLs”. Chatbit lists the pages it finds on that domain, and you add the ones you want with one click each.
- Search filters the results by a keyword, for example “pricing” or “blog”.
- Include subdomains also lists pages on subdomains such as
docs.example.com. - Ignore sitemap crawls the site instead of reading its sitemap, which helps when the sitemap is missing or incomplete.
The number of pages mapped per run depends on your plan. Start with the pages that answer real customer questions: FAQ, pricing, product, shipping and contact pages.
Files
Drag and drop files onto the “Add training data” box or click to pick them. Supported file types are .pdf, .doc, .docx and .txt. For best results, use files that do not rely on heavy design formatting. Any images or charts inside a document are not used.
Text
Type or paste free-form text into the same box. This is useful for information that is not on your website yet: opening hours, contact details, policies, or a short summary of what your company does.
Long text is split into smaller chunks. Click a source in the Training data list to see its chunks; you can edit or delete each chunk individually.
Q&A
Q&A sources are created from your chat logs. Open Dashboard, find a conversation with a weak answer and click “Revise Answer”. Write the answer you would have wanted and save it — it becomes a Q&A source, so the chatbot has the right answer the next time a visitor asks something similar. The original conversation stays unchanged.
We recommend reviewing the logs regularly and turning the questions your chatbot struggles with into Q&A sources.
Editing and removing sources
Click any source to expand its chunks. Edit a chunk to fix a mistake, or delete chunks that are irrelevant. Use the × icon to remove a whole source. The badge above the list shows how much of your plan’s character limit you are using; if you go over it, remove a source or upgrade your plan.
Next steps
After adding sources, we recommend proceeding with Model settings, where you can adjust your chatbot’s instructions and choose the AI model.
Next guide: Configuring model settings of your chatbot