How to Write Effective Prompts for AI Copilot

AI Copilot can turn natural language into an RPA workflow, but the quality of the workflow depends heavily on how clearly you describe what you want.

Clear, specific instructions help the AI Copilot understand exactly what you want, while vague or ambiguous prompts can lead to errors or incomplete workflows. By learning how to structure your prompts effectively, even users without prior RPA experience can create accurate and reliable automation flows.

In the following sections, we’ll guide you through practical strategies for writing prompts, explain common pitfalls, and show how step-by-step instructions can improve AI’s understanding and output.

You don’t need any RPA experience to write a good prompt, but it helps to know what you want the automation to achieve. Think about the task the same way you would perform it on a real webpage. For example, imagine how you would search for something, click a button, close a popup, or gather information from a page. When you understand the sequence of actions, it becomes easier to explain it to AI Copilot.

AI Copilot works best when you describe your idea step by step. Long paragraphs mixing clicks, searches, data extraction and page transitions tend to confuse the Copilot. Instead, write one action at a time and keep each instruction focused on a single intent.

If the page contains text labels such as “Add to Cart,” “Submit,” or “Next,” include those exact words in your prompt. When a button doesn’t have text and only shows an icon, describe what the icon looks like or where it sits on the page. The goal is to help the AI match your instructions to the correct element.

Try to avoid vague language such as “maybe,” “some,” or “kind of like.” AI Copilot will make better decisions when the instructions tell it exactly what to click, what to type, and what to extract.

Clear prompts tend to sound calm and direct. Here’s a simple example. A vague prompt might say: “Search for a product, open it, and get the details.” The AI can guess what you mean, but it has no sense of which search bar to use, which result to click, or what “details” means.

If you’d like to see what a full, well-structured prompt looks like from start to finish, here’s a teaching example. The website below is used purely for demonstration to show how a clear, step-by-step prompt is written:

Go to https://www.realtor.ca/.

In the Residential search box on the homepage, enter “Toronto, ON” and press Enter to perform the search.

Wait until the search results page fully loads.

On the results page, locate the filter bar and click the “Max Price” dropdown.

Select “75000” from the dropdown options.

Wait until the filtered results load completely.

For each property listing in the results:

Click the property to open its detail page.

Wait for the detail page to load fully.

Collect the following information:

Property price,Property address,Salesperson name and contact information,Property detail page URL.

Go back to the results page to continue with the next listing.

Save all the collected property data into an Excel file.

It breaks the task into clear steps, identifies exact actions on each page, specifies what data to collect, and explains where and how to save results. Step-by-step instructions, page-specific wording, and a predictable pattern for repeated actions help the AI understand loops and transitions, producing a more accurate workflow.

Even with a clear prompt, AI Copilot may occasionally click the wrong element or extract incomplete data. This doesn’t mean you need to rewrite everything. Often, adding one or two details when you check the preview will solve the issue. For example, if the AI clicks the wrong button, you can tell it to use the button with a specific label or location. Think of interacting with AI Copilot as a conversation. Each clarification teaches the model more about what you expect, and the workflow improves with every adjustment.

If you want a deeper understanding of how to structure your ideas before writing a prompt, the article How to Turn Your Automation Idea into an RPA Workflow is a helpful next step. That guide explains how to map out your logic, while this article focuses on how to communicate that logic to the AI. Used together, they give you a strong foundation for building better workflows with AI Copilot.