Many Zapier products allow you to interact with an AI (artificial intelligence) system using a prompt. This article explains how to create an effective prompt.
What is prompting?
In the context of AI, prompting is asking the AI to do something. You provide the AI with specific inputs, or information the AI needs to complete its task, and it will attempt to provide you with your ideal outputs, or results. The way you describe your requests and the information you provide the AI influences the quality of the results.
Where can I prompt AI within Zapier products?
The following Zapier products use prompts:
- Zapier Copilot.
- AI by Zapier.
- The AI Field in Zapier Tables.
- When creating a new canvas in Zapier Canvas.
- Zapier Agents.
- Zapier Chatbots directive.
Prompt structure
When creating your prompt, ensure that it:
Uses clear, plain language
- Be direct and concise.
- Avoid ambiguous words or phrases.
- Use simple sentence structures.
Is coherent and specific
- Your prompt should follow a logical structure. For example, write your prompt as step-by-step instructions.
- Ensure your instructions do not conflict or override each other.
- Describe all tools and fields you want the AI to use. For example, if you want your output to be added to a Zapier table, your prompt can say, “For each new order you analyze, create a new row in Zapier Tables for each item in the order.”
Provides context
- Describe the task you want the AI to perform, including all steps your AI needs to perform for your workflow.
- Include as much description and context as possible, including examples of both the inputs and outputs.
- Ensure you detail the most important information and steps to complete the task.
- It can be helpful to describe the role you want your AI to fulfill its tasks as. For example, if you want the AI to write a help article, you can include: “You are an experienced technical writer with advanced knowledge of best practices for writing for software applications.”
Describes your desired format
- Be specific about the style, length, and format of your desired output. For example, if you want the AI to return a summary in outline form, tell it: “Format the summary as an outline. Use a bulleted list for each item in the outline.”
- When possible, use bulleted or numbered lists to clarify what you want the AI to do or the output you want it to return.
Test and refine
After you create your prompt, test it. Depending on your result, you can continue to edit and refine your prompt to optimize it.
Context window
LLMs (large language models) use context windows, which determine the amount of information they can process at once. Window sizes are measured in tokens, or parts of a word, and depend on the AI model used. If your prompt has too many steps or requirements, it can become too complex to run effectively.
Depending on the AI tool you're using, you can:
- Give the AI small amounts of information at a time and test to ensure that the result is what you expect.
- In Zapier Agents, create separate agents to handle different parts of a complex request, then have one agent call the others to link their work.
You can use the Formatter app’s Split Text into Chunks for AI Prompts to break up your data into smaller chunks to use with AI apps.
Example use cases
Below you'll find examples of prompts for specific use cases. Use these as inspiration for your own prompts.
| Use case | Example prompt | Product |
|---|---|---|
| Summarize messages | Create a summary of all customer feedback received through my chatbot in the past month. Use Google Docs to create it, with the following name convention for the title: "Customer feedback - month name". Use one bullet point for each topic, and include two quotes from customers. Alert me in Gmail when it's ready. | Zapier Agents |
| Create prioritization system |
Create a system that receives tickets through Jira, and uses the “Summary” and “Description” fields of new tickets to understand the request and customer sentiment, and categorize them. Use the following options:
It should notify the IT team via Microsoft Teams whenever a new ticket has been categorized. |
Zapier Canvas |
| Compose an intro email | When a new record is added to my table of new leads, use the fields “Name”, “Company”, “Location”, and “Products selected” to compose a personalized welcome email to our business. | Zapier Tables |
Next steps
Creating a prompt that matches your needs takes some trial and error. Make as many changes as you need to get the AI to generate the best assets and steps for your use case.
You can also check out specific information on how to create prompts for Zapier products:
- Prompt AI to create a system in Zapier Canvas.
- Improve an agent's behavior results.
- Create a directive and greeting for your chatbot.
Learn more: