AB-410 Tutorial: Build an AI Prompt and Use It in Power Apps and Power Automate (2026)

Last updated: October 2026. Fact-checked by the Examinotion editorial team against Microsoft Learn exam pages, the AB-410 study guide and current prompt builder, AI Builder and Power Platform documentation.

TL;DR: To pass the AB-410 prompts objectives, build one prompt end to end: create it in AI hub, add a text input and Dataverse knowledge, switch the output to JSON, then call it from Power Automate with the Run a prompt action and from a canvas app with Predict. Testing in prompt builder is free; running it consumes credits.

Exam AB-410: Building Intelligent Applications leads to the Microsoft Certified: Intelligent Applications Builder Associate certification [2], and its largest domain, Build business application logic and automation, carries 40-45% of the marks [1]. Inside that domain sits a whole skill group called Create prompts and models in AI Hub, with eight testable skills [1]. This tutorial walks you through building a single, realistic prompt and consuming it in a cloud flow, a canvas app and a Dataverse prompt column, so you have hands-on experience of every prompt skill the exam measures.

What you will build, and which AB-410 skills it covers

The tutorial builds a support ticket triage prompt. It takes the text of a customer ticket, looks up relevant product records in Dataverse, and returns a structured JSON response with a category, a priority and a one-sentence summary. You then use that one prompt in three places, because AB-410 tests consumption in apps and in flows as separate skills [1].

The AB-410 study guide lists these skills under Create prompts and models in AI Hub [1]:

The study guide writes the area as "AI Hub", while the product navigation labels it "AI hub"; both refer to the same place. Two neighbouring skills also benefit. Configure prompt columns sits in the Build data models group of the foundation domain, and Configure cloud flow actions sits in the Create cloud flows group [1]. The study guide's prerequisites explicitly expect "foundational knowledge of AI models and prompts in AI Hub" [1], so this is not optional reading for AB-410 candidates.

For the full exam structure, read our AB-410 exam guide or the official AB-410 study guide on Microsoft Learn.

Prerequisites: environment, Dataverse and credits

A prompt needs a Power Platform environment with Microsoft Dataverse, in a supported region, with Copilot Credits available [5]. The usual low-cost way to get a practice environment is the Power Apps Developer Plan, which Microsoft describes as "a free development environment to build and test applications with Power Apps, Power Automate, and Microsoft Dataverse" [21]. Sign-up requires a work or school account backed by Microsoft Entra ID; personal accounts cannot sign up [21].

Credits are the part most candidates misunderstand. Microsoft states that building and testing prompts does not require AI Builder credits or Copilot Credits, but running them in agents, agent flows, apps or flows does consume credits [17]. Microsoft also states that AI Builder trials are discontinued [17]. In practice, you can complete Steps 1 to 5 of this tutorial at no cost, and you need either AI Builder credits or Copilot Credits in the environment before the flow, app and prompt column steps will run successfully. A Developer Plan environment does not come with AI Builder use rights [21], so for those steps you may need an environment that your organisation has provisioned with credits.

Before you start, confirm four things:

1. Dataverse is provisioned in the environment (Developer Plan environments include it) [21].
2. The AI prompts feature is switched on. The Power Platform admin centre setting "AI prompts" enables the feature in Power Platform and Copilot Studio and is on by default [6].
3. You have a Dataverse table to ground on. Create a simple Product table with columns for product name, support tier and known issues, and add five or six rows. Any table you own works.
4. You have a Ticket table for the app and flow steps. Create a Ticket table with a multiline Description column plus three text columns, Category, Priority and Summary, which the flow will fill from the prompt's output.

Step 1: Create a prompt in AI hub

Prompts are created in the prompt builder, which Microsoft describes as an experience that "allows makers to build, test, and save reusable prompts" [5]. Prompt documentation now lives under Microsoft Copilot Studio, but the same prompt builder opens from Power Apps and Power Automate.

1. Sign in to Power Apps (make.powerapps.com) or Power Automate.
2. In the left navigation, select AI hub, then Prompts, then Build your own prompt [6].
3. Give the prompt a clear name, such as Ticket triage. You will call it by this name from Power Fx later, so keep it short.
4. Alternatively, start from a template in the prompt library. The exam skill explicitly covers building prompts "from templates or from blank" [1], so open the template list once and note how a template pre-fills instructions.

Write the instructions in plain language. A workable starting instruction is: "Read the support ticket. Classify it into one category: Billing, Technical, Account or Other. Set priority to High, Medium or Low. Write a one-sentence summary. Use the product records to identify the product the customer mentions."

New prompts are private by default [20], so share the prompt with colleagues explicitly if they need to reuse it.

Step 2: Add inputs to a prompt

An input is a placeholder the calling app or flow fills at run time. In the Instructions panel, type a forward slash or select Add content, then choose the input type: Text, Image or document, or Power Fx [7].

For this tutorial, add one Text input named Ticket text and give it a realistic sample value, such as a short complaint about a failed card payment. The sample value is only used when you test inside prompt builder.

Know these input rules for the exam, because they make good distractors:

• Image or document inputs accept PNG, JPG, JPEG and PDF, plus Word, Excel and PowerPoint files when code interpreter is turned on in prompt settings; all files together must be under 25 MB, and documents must have fewer than 50 pages [7].
• Image or document processing is limited to 100 seconds [7].
• "Inputs can't contain instructions" [7]. Put instructions in the prompt, not in the data passed to it.
• Image or document input is not yet supported for prompts added as tools in Copilot Studio agents [7].
• A Power Fx input is evaluated as a formula, and can reference other inputs with the syntax Input.'Your input name' [7].

Step 3: Ground the prompt with Dataverse knowledge

Knowledge lets the prompt retrieve records at run time instead of relying only on the model's general training. In the Instructions panel, insert knowledge and select your Product table, then choose the columns the prompt should read and, optionally, filter attributes.

The knowledge feature has firm boundaries, and AB-410 scenarios frequently probe them. Microsoft's limitation note reads: "Data sources are limited to Dataverse and Salesforce, Oracle, and ZenDesk connector tables" [8]. The same page's connector section also names SAP alongside Salesforce, Oracle and Zendesk [8]. Other limits from that page [8]:

• Connector-based knowledge can only be used for prompts in Power Automate.
• Virtual tables are not yet supported.
• For Dataverse, only one-to-many and many-to-one relationships are supported, up to two levels deep.
• The prompt retrieves 30 records by default, and you can raise this to 1,000 in settings.

A frequent exam trap is to offer SharePoint or uploaded files as prompt knowledge. That is grounding for a Copilot Studio agent, not for a prompt built in AI hub. If a question asks how to ground a prompt on business data held in Power Platform, Dataverse is the answer to look for.

Step 4: Choose the model and settings

Open the three-dot menu in prompt builder and select Settings to change the model and behaviour. The default model is GPT-4.1 mini, billed at the basic rate [9]. Other options include GPT-4.1, GPT-5 chat and GPT-5 reasoning, plus several models Microsoft labels as experimental [9]. Experimental models should not be treated as production-ready.

The settings panel includes [9]:

Keep temperature at 0 for classification tasks like triage, because you want the same ticket to produce the same category every time. Region matters too: in the United Kingdom, GPT-4.1 mini and GPT-4.1 are generally available in region, while GPT-5 chat and GPT-5 reasoning are available cross-geo, meaning data might be processed outside your region [10].

Step 5: Return JSON output and test

By default a prompt returns text [11]. To make the response usable by a flow or app, select JSON as the output in the top-right corner of prompt builder [11]. The format starts as Auto detected, which refreshes each time you test; if you edit the JSON example, the format becomes Custom and stops updating [11]. Saving the prompt locks whichever format is current [11].

A suitable JSON example for the triage prompt is:

{
  "category": "Billing",
  "priority": "High",
  "summary": "Customer's card payment failed twice on the annual plan."
}

Select Test with your sample ticket and check that the response matches the structure. Two JSON rules are worth memorising [11]:

• You cannot modify a JSON schema directly; you shape it through the example.
• Every value needs a field key. An array of bare strings is not supported, whereas an array of objects with keys is.

If testing returns the error "A JSON could not be generated", Microsoft's documented fix is to add the instruction "Don't include JSON markdown in your answer" [11]. Testing inside prompt builder is free [17], so iterate here until the output is consistent, then save.

Step 6: Consume the prompt in a Power Automate cloud flow

The Power Automate action that calls a prompt is Run a prompt. Microsoft notes that "Since May 2025, Create text with GPT using a prompt action name is Run a prompt" [12], so older study material using the previous name is out of date.

1. Create an automated cloud flow with the Dataverse trigger When a row is added, modified or deleted, set to fire when a row is added to your Ticket table.
2. Add the Run a prompt action and select Ticket triage in the Prompt field [12]. You can also choose New custom prompt to build one inline [12].
3. Map the Ticket text input to the ticket description column from the trigger.
4. Add an Update a row action for the same ticket, and fill the Category, Priority and Summary columns from the dynamic content the prompt exposes.

For text output, the action returns a Text variable you can drop into a later step [12]. For JSON output, Microsoft's worked example inserts the extracted fields directly as dynamic content in the next action [11]. The documentation does not describe a separate Parse JSON step for this pattern, so start with the fields the action exposes.

Two operational facts appear in licensing and scenario questions. A flow that uses any AI Builder action, including Run a prompt, is a premium flow [17]. And every run consumes credits, including a system metaprompt of roughly 1,200 tokens on top of your own instructions and inputs [17]. For human oversight on sensitive outputs, pair the prompt with an approval action before the update.

Step 7: Consume the prompt in a canvas app

Canvas apps call a prompt as a Power Fx function. In Power Apps Studio, open Data, select Add data, and choose your prompt [13]. Once added, "it can be used as a Power Fx function by calling the .Predict on its name" [13].

Microsoft's documented pattern places the call on a button and stores the result in a variable [13]:

Set(result, 'Ticket triage'.Predict(TextInput1.Text));

Set a label's Text property to result.text to show a text response [13]. If your prompt returns JSON, inspect the result variable in the formula bar to see the record the prompt returns before you bind fields to controls, rather than assuming its shape.

Trigger predictions from a single deliberate action. Microsoft recommends calling a model from an action such as a button's OnSelect rather than OnChange, to ensure efficient use of AI Builder credits [14]. In practice, an OnChange trigger can call the prompt far more often than the user intends. Adding an AI Builder action to an app also makes it a premium app for licensing purposes [19].

The model-driven alternative: prompt columns

Model-driven apps usually consume prompts through prompt columns rather than formulas. Microsoft defines a prompt column as "an AI-powered data type in Microsoft Dataverse that enables you to define natural language prompts tied to other columns in your table" [15]. Prompt columns became generally available in July 2026 [16], so treat them as a GA feature in exam scenarios.

To add one, open your Ticket table, create a new column, set the data type to Prompt, and select Add new prompt [15]. Key behaviours to remember [15]:

• A table can have up to five prompt columns.
• The prompt runs when a record is created or when a referenced column is updated.
• Existing records are not processed by default, so there is no automatic backfill.
• Input columns cannot be formula columns, file columns, image columns or other prompt columns.
• Prompt columns consume AI Builder credits and Copilot Credits.

Add the prompt column to the main form and it appears in any model-driven app that uses that form. This maps to the Configure prompt columns skill in the foundation domain [1].

Credits and licensing in 2026

Prompt billing is changing during 2026, and outdated material is common. Microsoft's current position is that AI Builder features in Power Apps and Power Automate consume either AI Builder credits or Copilot Credits, while agents and agent flows consume only Copilot Credits [17].

AI Builder credits are being phased out in two stages [18]:

Where AI Builder credits remain, they are consumed first; if they run out, the system attempts to use Copilot Credits, and if none are available, usage is blocked [18]. Prompts always consume credits, even when the selected model is in preview [19]. For exam purposes, the safe summary is that testing in prompt builder is free and running a prompt from a flow, app or agent costs credits [17].

Responsible AI, data handling and common errors

Prompt builder prompts use Azure AI Content Safety, including protection against prompt injection attacks, and Microsoft states that customer data is not used to train or improve the Azure OpenAI Service foundation models [20]. The content moderation level, from Low to High with Moderate as the default, is configurable for managed models [9].

Data location is an environment-level decision. In the Power Platform admin centre, the Generative AI features card for an environment includes a Move data across regions option, and Microsoft notes that inputs and outputs "might move outside of your region" when it is enabled [22]. UK organisations with data residency requirements should check this setting before rolling out prompts.

The errors and design faults candidates meet most often:

• "A JSON could not be generated": add the instruction "Don't include JSON markdown in your answer" [11].
• Predictions firing too often: move the Predict call from OnChange to a button's OnSelect [14].
• Instructions in the input: inputs cannot contain instructions [7].
• Image input in an agent tool: not yet supported for prompts used as Copilot Studio agent tools [7].
• Flow or app suddenly premium: expected behaviour once an AI Builder action is added [17][19].

How this maps to the AB-410 exam

The prompt skills sit inside the domain worth 40-45% of the exam, which makes them some of the highest-value hands-on practice you can do [1]. The exam lasts 120 minutes, requires 700 or more to pass, and may include interactive components [1][2]. As an associate-level exam, it falls under Microsoft's general policy that allows access to Microsoft Learn during associate and expert exams [4].

AB-410 is not an exam you can pass from reading alone. Scenario questions often hinge on small distinctions: where prompt knowledge can come from, which action name is current, whether a feature is GA, and what consumes credits. Microsoft notes the exam "may contain questions on Preview features if those features are commonly used" [1], so practising in a real environment is the most reliable way to build that judgement.

AB-410 is also the closest onward path for Power Platform functional consultants. Microsoft's August 2026 partner announcement states that PL-200 retired on 31 August 2026 and was replaced by the Intelligent Applications Builder Associate (AB-410), available from 30 July 2026 [3]. Read our PL-200 retirement and AB-410 successor guide if you are making that move, and see the Microsoft AI certification roadmap for where AB-410 sits among the other AI certifications.

Next steps: practise under exam conditions

Once the triage prompt works in a flow, a canvas app and a prompt column, extend it to test the edges: add an image input for screenshots, raise record retrieval, switch the model, and watch how output and credit consumption change. Then check your understanding with the worked items in our AB-410 practice questions and answers.

If you are deciding between low-code app building and agent building, compare this tutorial with the agent content in our AB-620 exam guide and Microsoft Copilot agents business guide. For structured revision across every AB-410 domain, use the AB-410 study guide.

Frequently asked questions

Where do you create a prompt for Power Apps and Power Automate?

You create a prompt in prompt builder. Sign in to Power Apps or Power Automate, open AI hub in the left navigation, select Prompts, then Build your own prompt. The same prompt builder is available from Microsoft Copilot Studio, and a saved prompt can be reused across apps, cloud flows and agents once it has been shared.

What is the Power Automate action for running an AI prompt?

The action is called Run a prompt. Microsoft renamed it in May 2025 from Create text with GPT using a prompt, so older study material may still use the previous name. You select the saved prompt in the action, map each input to dynamic content, and use the returned text or JSON fields in later steps of the cloud flow.

What data sources can an AI prompt use as knowledge?

Prompt knowledge centres on Dataverse tables, plus connector tables from business systems such as Salesforce, Oracle, SAP and Zendesk, and connector knowledge only works for prompts used in Power Automate. Virtual tables are not supported, Dataverse relationships go two levels deep, and retrieval defaults to 30 records, which you can raise to 1,000.

How do you call a prompt from a canvas app?

Add the prompt to the canvas app through Data and Add data, then call it as a Power Fx function using Predict on the prompt's name. Microsoft's example sets a variable from a button's OnSelect property and displays the text result in a label. Trigger the call from a deliberate action rather than OnChange to avoid wasting credits.

Does testing a prompt in prompt builder cost credits?

No. Microsoft states that building and testing prompts does not require AI Builder credits or Copilot Credits. Credits are consumed when the prompt runs inside an app, cloud flow, agent or agent flow. AI Builder trials are discontinued, so make sure your practice environment has AI Builder credits or Copilot Credits before testing your flow and app steps.

Are prompt columns in Dataverse generally available?

Yes. Microsoft announced in July 2026 that prompt columns in Microsoft Dataverse are generally available, after a public preview that began in August 2025. A table can hold up to five prompt columns, prompts run when a record is created or a referenced column changes, and existing records are not processed automatically, so plan any backfill separately.

Is AB-410 hard if you have never used Power Platform?

AB-410 is an intermediate, associate-level exam with a 700 pass mark, and candidates without Power Platform experience usually find it demanding. Nearly half the marks come from business logic and automation, which reward hands-on practice with cloud flows, Power Fx and prompts. Building small solutions in a practice environment before booking makes the scenario questions much more manageable.

Conclusion

Building one prompt and consuming it three ways covers every prompt skill in the AB-410 study guide: templates and blank prompts, inputs, knowledge, settings and models, and consumption in cloud flows, canvas apps and prompt columns. Keep the exam-sensitive details front of mind: the Run a prompt action name, the Dataverse-centred knowledge limits, JSON format rules, premium licensing and the 2026 move from AI Builder credits to Copilot Credits.

When you are ready to test that knowledge under timed conditions, start practising for AB-410 with Examinotion's AB-410 practice tests, or browse all of Examinotion's Microsoft exam preparation courses. Preparation reduces risk on exam day; it does not guarantee a pass, so book once you are consistently clearing practice scores comfortably.

Sources

1. Study guide for Exam AB-410: Building Intelligent Applications - Microsoft Learn, accessed 2026-10-12
2. Microsoft Certified: Intelligent Applications Builder Associate - Microsoft Learn, accessed 2026-10-12
3. Partner Center announcements, August 2026 - Microsoft Learn, accessed 2026-10-12
4. Exam duration and exam experience - Microsoft Learn, accessed 2026-10-12
5. Prompts overview - Microsoft Learn, accessed 2026-10-12
6. Create a prompt and Manage feature settings - Microsoft Learn, accessed 2026-10-12
7. Add inputs to a prompt - Microsoft Learn, accessed 2026-10-12
8. Use your own data in a prompt - Microsoft Learn, accessed 2026-10-12
9. Prompt model settings - Microsoft Learn, accessed 2026-10-12
10. Prompt model availability - Microsoft Learn, accessed 2026-10-12
11. Process responses with JSON output - Microsoft Learn, accessed 2026-10-12
12. Use your prompt in Power Automate - Microsoft Learn, accessed 2026-10-12
13. Use your prompt in Power Apps - Microsoft Learn, accessed 2026-10-12
14. Use Power Fx in Power Apps with AI Builder - Microsoft Learn, accessed 2026-10-12
15. Prompt columns in Dataverse - Microsoft Learn, accessed 2026-10-12
16. Prompt columns in Microsoft Dataverse are now generally available - Microsoft Power Platform Blog, accessed 2026-10-12
17. Administer AI Builder licensing and Prompt token consumption - Microsoft Learn, accessed 2026-10-12
18. End of AI Builder credits - Microsoft Learn, accessed 2026-10-12
19. AI Builder credit management - Microsoft Learn, accessed 2026-10-12
20. Prompts FAQ - Microsoft Learn, accessed 2026-10-12
21. Power Apps Developer Plan - Microsoft Learn, accessed 2026-10-12
22. Geographical availability of Copilot and generative AI features - Microsoft Learn, accessed 2026-10-12