If you’ve been working with AI in your business processes lately, you’ve probably run into a really frustrating roadblock: the gap between the chatty text an AI wants to give you and the structured data your automation actually needs.

Let’s say you’re building a Power Automate flow to pull details out of customer emails. You want the person’s name, the priority, and the category. Ideally, you want to get back something clean like this:

JSON:

{

"customer_name": "Sarah Johnson",

"priority": "High",

"issue_category": "Billing"

}

Instead, the AI usually gives out a whole paragraph:

“The customer, Sarah Johnson, appears to have a high-priority concern related to billing. It would be advisable to route this to the finance team for further assistance.”

That’s great if a human is reading it. But for an automated flow? It’s a nightmare. You end up trying to write expressions using ‘split()’ or regular expressions just to grab the values you want, and if the AI rephrases even slightly on the next run, those expressions break. The core issue is straightforward: AI produces conversational text, but automated systems require structured, predictable data.

Microsoft’s Power Platform has a fix for this inside the AI Builder Prompt action. Whether you are using Power Automate or Copilot Studio, you can just tell the AI to hand you back perfectly formatted JSON instead of a paragraph. And the best part? That JSON maps straight into your Dynamics 365 fields, your flow conditions, or your Copilot variables. No messy parsing needed.

Understanding AI Builder Prompts

The Prompt tool can be accessed via the Power Apps maker portal, under AI Builder in Prompts. It allows users to define a custom AI instruction with optional dynamic input variables, and the generated response can be consumed in Power Automate or in Copilot Studio. By default, the output is plain text, but switching the output format to JSON enables the AI to return a structured object that downstream actions consume directly.

As a general guideline: use text output when communicating with a user, and JSON output when communicating with a system.

Setting Up a Prompt with JSON Output:

Navigate to Power Apps maker portal, under AI Builder in Prompts select the Build your own Prompt.

When writing the prompt, three elements are essential: a clear task definition, the JSON schema (field names, data types, allowed values), and explicit output constraints (return only the JSON, no extra text or formatting). Define dynamic inputs by clicking on Input, for example, an ‘EmailBody’ variable to hold customer email text passed from a flow.

In the configuration panel, change the Output format from Text to JSON. Use the built-in Test feature to validate the response structure.

Once tested, integrate it into Power Automate via the “Run a prompt” action, or into Copilot Studio as a Prompt Tool directly or use it within a topic.

Automated Email Categorization with Power Automate

Support teams get flooded with emails every single day. Usually, a human has to sit there, read each one, figure out the priority, guess the customer’s mood, and write a summary.

By combining the Prompt tool with JSON output, you can put this whole categorization process on autopilot. Here is roughly how it flows:

– An Email Arrives: A distressed customer sends an email regarding a critical payment failure on their account.

– The Prompt Analyzes: Your flow triggers immediately, grabs that email text, and ships it over to the AI Builder Prompt you built.

– Clean Data is Structured: The AI reads it, understands the context, and hands you back a neat JSON block.

– Parsed for Automation: Using the “Parse JSON” action in Power Automate, the structured data is instantly converted into dynamic content. You can now use these variables anywhere in your flow to route the email to the right team, trigger an urgent alert, or seamlessly create a record in Dataverse.

Conclusion

Switching AI Builder Prompts to JSON output is a simple but powerful enhancement for building production-ready automation.

While text responses are useful for human interaction, JSON output provides the structure required for reliable system integration.

This approach:

• Eliminates fragile parsing logic
• Ensures consistent AI responses
• Enables direct mapping to business systems

For any solution involving Power Automate, Copilot Studio, or Dynamics 365, using JSON output is one of the most impactful improvements you can make with minimal effort.

FAQs

What is JSON output in AI Builder Prompts?

JSON output is a structured response format in AI Builder Prompts that returns data as key-value pairs instead of conversational text, making it easier for automation systems to process.

 Can JSON output integrate with Dynamics 365?

Yes, structured JSON fields can map directly into Dynamics 365 entities and records without additional transformation.

 Can JSON output integrate with Dynamics 365?

Yes, structured JSON fields can map directly into Dynamics 365 entities and records without additional transformation.

What problems does JSON output solve in AI automation?

It eliminates inconsistent AI responses, removes the need for regex or string parsing, and ensures stable automation workflows.

How do I enable JSON output in Power Apps AI Builder Prompts?

In the Power Apps maker portal, go to AI Builder → Prompts, create or edit a prompt, and change the output format from Text to JSON in the configuration panel.

The post How to Use JSON Output in AI Builder Prompts for Structured Automation first appeared on Microsoft Dynamics 365 CRM Tips and Tricks.

In today’s fast-paced digital world, automating repetitive tasks not only saves time but also significantly improves productivity. Microsoft now offers a powerful preview AI model that enables automatic generation of image descriptions using the AI Builder’s prebuilt Image Description model in Power Automate Flow.

This smart tool analyzes your images and generates easy-to-understand, meaningful descriptions. These descriptions are really helpful for organizing your files, sorting images, and making content more accessible, all without you having to do anything manually.

How the AI Builder Image Description Model Works

The AI Builder Image Description model uses advanced computer vision to understand what’s in an image and convert that visual content into meaningful text. This makes tasks like organizing files, generating metadata, creating reports, or improving accessibility much easier because the system provides clear descriptions instantly and automatically.

Here’s a breakdown of how the model works behind the scenes:

1. It analyzes the image and generates three key outputs:

• A description (English only): A simple, human-readable explanation of what the image contains.
• Tags: Keywords that highlight the main objects, themes, or concepts detected in the image.
• A confidence score: A percentage indicating how certain the model is about its description.

2. It supports only these image file formats:

.JPG, .JPEG, .PNG, and .BMP
Uploading any other format will cause the action to fail.

3. Image size requirements:

1. • Maximum file size: 4 MB
   • Minimum resolution: 50 × 50 pixels

4. Role and licensing requirement:

You only need the Basic User role to use this model inside a Power Automate flow with no special admin permissions required.

Important note:
This feature is currently in preview, which means it works reliably for simple descriptions but is not recommended for production use yet.

Prerequisites:

Before creating a flow using this model, ensure you have:

• Access to Microsoft Power Automate
• Access to Dataverse

Step-by-Step Guide: Automate Image Descriptions

1. Create a New Flow

• Sign in to your Dataverse
• Open Power Automate, and click Create → Instant cloud flow (or choose another flow type based on your use case).

2. Add File Input (Optional)

• In the trigger step, click Add an input → Select File.

• This allows you to upload an image manually when testing the flow.

3. Add the AI Builder Action

• Click New Step → Search for “AI Builder” → Select “Describe images (Preview)”.

• In the Image field, choose File Content from Dynamic Content.

4. Add Post-Processing Logic (Optional)

• Click once the description is generated, you can add further steps such as:
  • Storing the description in a database.
  • Sending an email notification.

5. Save and Test the Flow

• Click Save.
• Select Test → Manually, and upload an image when prompted.
• The flow will run and automatically generate a description, confidence score, and related Tags.

Important Consideration (as per Microsoft docs):
This AI Builder currently supports only the English language and the following image formats only: Jpeg, Png, Gif, Bmp. Uploading other file types will result in a failed operation.

This feature is still in preview and currently provides straightforward descriptions. However, in the future, it has the potential to generate more detailed and complex image descriptions.

When Should You Use the Image Description Model?
A common use case is when a user uploads product images to your system. Automatically generating a description helps:

• Product Image Metadata: Automatically generate captions for product images in catalogs.
• Accessibility: Provide alt-text for images on websites and in documents.
• Content Tagging: Tag images with relevant labels for smarter search.
• Surveillance or Monitoring: Describe visual scenes (people, objects, activity) for easier review or alerts.

FAQs

1. How does the AI Builder Image Description model help my workflow?
   It automatically scans your images and generates clear, meaningful descriptions along with tags and confidence scores. This means faster content organization, better accessibility, and zero manual effort.
2. What image formats can I upload?
   The model currently supports JPG, JPEG, PNG, and BMP files. Using any other format will stop the flow from running successfully.
3. Are there any image size limits I should know about?
   Yes, your image must be under 4 MB and at least 50 × 50 pixels. Staying within these limits ensures smooth processing.
4. Does it work with multiple languages?
   For now, it generates descriptions only in English, as highlighted in the blog.
5. Is this ready for production use?
   Not yet. Since the feature is still in preview, it’s ideal for testing, prototyping, and internal automation but not for mission-critical production scenarios.
6. Do I need admin rights to use this in my Power Automate flow?
   No. The Basic User role is all you need to start using the Image Description model, making it easy for anyone in your team to adopt.

Conclusion:
Power Automate’s Image Description prebuilt model makes it effortless to generate meaningful image descriptions — without writing a single line of code. Whether it’s for improving accessibility or automating content organization, this tool empowers you to streamline processes, save time, and increase efficiency.

The post How to Automate Image Descriptions with AI Builder in Power Automate first appeared on Microsoft Dynamics 365 CRM Tips and Tricks.