Use knowledge sources with a chatbot

Beta

This product is in open beta. It’s available for use but still in active development and may change.

You can connect Zapier Chatbots to your knowledge sources to ensure responses are focused on your business or project. The chatbot can use files or tables as its exclusive source of information for answers, and you can also set a custom response for when information cannot be found in the file.

Note

Zapier Chatbots was created to answer straightforward questions that your customers or users may have, such as "Do you deliver to Sydney, Australia?" and other information retrieval tasks. If you need an analysis or want to gain insights into a file's contents, try using Zapier Agents.

Add knowledge sources

Knowledge sources can be files, webpages or table using Zapier Tables. 

Add a webpage Add files Add from Zapier Tables

Use a webpage

You can add a webpage as a knowledge source, and the chatbot will scrape the content and use it to answer questions. 

1. From the Chatbots home, click the name of the chatbot.
2. Click the Knowledge tab.
3. Click + Add source.
4. On the Type section, click Webpage. A dialog box will open.
5. Paste an address in the Public URL field.
   • Optional: Select the Crawl this link for all subpages checkbox to choose URLs from a list of available subpages.
6. Click Continue. It may take a moment to sync.

If you select the checkbox Crawl this link for all subpages, a list of available subpages will be displayed. Select the checkbox beside each page you want to use as a knowledge source.

The number of pages you can select depends on your plan limits and the number of other knowledge sources already connected. The maximum number is displayed at the top of the dialog box.

Note

Checking the Select all checkbox will select as many pages as possible, within the number of available knowledge sources. The pages will be selected in the order that they appear in the dialog box, from top to bottom.

Limitations 

• The chatbot can only use publicly available pages. It cannot log in to a website.
• Images and videos cannot be crawled.

Add files

When you use a text-based file, the chatbot will scrape its content and use it to answer questions.

1. From the Chatbots home, click the name of the chatbot.
2. Click the Knowledge tab.
3. Click + Add source.
4. The Type section will have File selected by default. Click Browse files.
5. Find and select the source file you want to use. 
6. Click Add source. It may take a moment to process, depending on the file size.

Limitations

• Maximum file size: 2MB.
• Accepted file formats: .txt, .csv, .json, .pdf, .doc, .docx, .pptx.
• Maximum number of files: the number of files depends on your Chatbots plan.

Use Zapier Tables

You can add a Zapier table as a knowledge source and sync it either automatically or manually to pick up changes.

1. From the Chatbots home, click the name of the chatbot.
2. Click the Knowledge tab.
3. Click + Add source.
4. On the Type section, click Table.
5. Click the search field and start typing the name of the table.
   • Click Create a new Table to start a new table and use it as a knowledge source.
6. Click the name of the table.
7. Click Add source. It may take a moment to sync.

Limitations

• Chatbots only use the first 10,000 records of your table and up to 50 columns.
• Only tables you own are available as knowledge sources for your chatbots.

Sync your knowledge source

When you add Zapier or a webpage as knowledge sources, the chatbot will collect the information available at that time. If the knowledge source changes, you can sync them manually or schedule when they should be updated. Learn how to sync and set up a schedule for your knowledge sources.

Delete a knowledge source

To delete a knowledge source:

1. In the Knowledge tab of your chatbot, click the Delete knowledge source icon , which will appear in the same field as the knowledge source name.
2. Click Delete to confirm you want to remove the data source. It might take a moment to remove it.

Structure your files

Zapier Chatbots breaks knowledge source information into smaller chunks, so the chatbot can efficiently use the most relevant pieces of information during a conversation. Because of this, when using files, the most effective way to structure your knowledge source is to group related information together. See below how the chatbot will use information in different knowledge source types:

• Structured data (such as CSV, JSON and Zapier Tables): the chatbot assumes information in the same row is related (this includes multiple columns). So information in cells A1, B1, and C1 would be grouped as a related chunk of knowledge for the chatbot.
• Unstructured file (such as PDF, DOC and TXT): the chatbot looks for a double return space to indicate separate pieces of information. Placing related information within a paragraph together with the header can help improve answer accuracy.

Adjust chatbot behavior for missing answers

If the chatbot uses knowledge sources, you can define how it will respond when the sources do not contain answers that match a specific question. 

To change the missing answer behavior:

1. In the Knowledge tab of your chatbot, click the When results are not found from the knowledge sources dropdown menu.
2. Select an option from the dropdown menu:
   • Generate an AI response without knowledge source (default): the chatbot will generate a response using the connected OpenAI model. It will not use any additional context provided by any connected knowledge sources. 
   • Show a custom message: write a message that will be displayed if the chatbot cannot find an answer within any connected knowledge sources. 

Tip

You can use a custom message to send users to other sources of help, such as your contact form. 

Data retention

Zapier's Data Retention, deletion, and export practices vary depending on the product. Learn more about the specific guidelines for the product you're using.

Provide feedback and get help

Submit a feature request, provide feedback on existing features, or get help from the Zapier Chatbots team.