Domain 3 of 3 · Chapter 2 of 3

Create prompts and models in AI Hub

Unlock the complete study guide.

Bundled into the existing Building Intelligent Applications premium course — no separate purchase.

14-day refund guarantee on one-time purchases and your first subscription charge.

Included in this chapter:

  • Prompts and AI models in AI Hub
  • Build a prompt from a template or from blank
  • Prompt inputs: text, files, and Power Fx
  • Ground a prompt with knowledge
  • Choose the model a prompt runs on
  • Tune prompt settings
  • Text or JSON output
  • Prebuilt and custom AI models
  • Share prompts and AI models
  • Call prompts and AI models from a canvas app
  • Choose a text or image model by its output
  • Run a prompt in a cloud flow
  • Use AI Builder models in a cloud flow
  • Prompt and AI model configuration by requirement

Prompt vs prebuilt AI model vs custom AI model

Decision pointPromptPrebuilt AI modelCustom AI model
What you supplyA natural-language instruction, inputs, and optional knowledgeNothing to train; the model is ready to useYour own training data, such as sample documents, labeled text, or images
Fits whenThe task is described in words: summarize, draft, rewrite, or answer from recordsThe task is common across businesses: receipts, language, sentiment, text recognitionThe fields, categories, entities, or objects are unique to your business
Before first useTest with sample input values, then saveNone; add it and call itTrain, then publish the trained version
Canvas app callAdd data, then 'Prompt name'.Predict(...)Add data > AI models, then Predict, or a prebuilt componentAdd data > AI models, then Predict, or the form processor or object detector component
Cloud flow callRun a promptThe model's AI Builder action, such as Detect the language being used in the textThe model's AI Builder action, such as Process documents
What comes backText (default) or the saved JSON formatThe model's fixed output fields, such as Language and ConfidenceYour trained fields, classes, or objects, with confidence scores

Decision tree

Task described in words:summarize, draft, rewrite?YesNoMust it use yourDataverse records?Working with dataunique to yourbusiness?YesNoYesNoPrompt withknowledgePromptCustomAI modelPrebuiltAI modelThen: share a prompt or custom modelas User; call Predict (app) or anaction (flow). Generic predictionmodels: scheduled runs only

Cheat sheet

  • Templates are editable starting points for saved prompts
  • Effective prompts supply both a task and relevant context
  • Audience and output constraints belong in the instructions
  • Prompt tests exercise sample inputs before integration
  • Apps and flows consume saved prompts rather than the prompt library
  • Prompt builder requires Dataverse in a supported environment region
  • Prompt execution depends on available capacity as well as a valid definition
  • Named text inputs make prompts reusable for changing content
  • File and text inputs can be combined in one prompt
  • Office file analysis requires code interpreter
  • File inputs have independent size, page, and processing limits
  • Accepted documents can still produce incomplete extraction
  • Power Fx inputs compute context before generation
  • Runtime inputs carry data rather than trusted instructions
  • Knowledge filter attributes support a defined set of data types
  • Knowledge grounding supplies organization-specific context
  • Knowledge objects identify the data columns used by a prompt
  • Knowledge filters can bind to prompt inputs
  • Dataverse knowledge supports limited relationship traversal
  • Record retrieval limits the data supplied as knowledge
  • Connector-based knowledge is restricted to Power Automate consumption
  • Prompt settings can include links to retrieved records
  • Model choice balances reasoning, latency, and consumption
  • Temperature changes variability rather than guaranteeing correctness
  • GPT-5 reasoning does not expose the temperature slider
  • Content moderation controls filtering rather than creativity
  • Preview and experimental models require evaluation rather than assumed production readiness
  • The model context window constrains each invocation
  • JSON output exposes separate elements for downstream actions
  • Custom JSON examples preserve structure across authoring tests
  • Saving fixes the output format used at runtime
  • JSON configuration uses keyed examples instead of direct schema editing
  • Conflicting instructions can prevent valid JSON generation
  • Canvas apps call added prompts through Predict
  • Run a prompt maps flow data into prompt inputs
  • The flow action provides an entry point for custom prompt authoring
  • Prompt sharing separates use from editing privileges
  • Text approval preserves a reviewer's accepted revisions
  • Generated content needs human review before external or consequential use
  • AI models support function calls and purpose-built app components
  • Available AI model data sources depend on model access
  • Explicit invocation and result reuse avoid unnecessary predictions
  • Language detection returns an identifier and confidence
  • Classification labels text while entity extraction returns values inside it
  • Sentiment and key phrases answer different analysis questions
  • Text recognition reads content while object detection locates trained objects
  • AI Builder actions should retain Asynchronous Pattern enabled
  • Process documents requires a published model and file content
  • A document page range should contain one complete form
  • Extracted values and confidence are distinct outputs
  • Extracted document tables require row-level processing
  • Extracted text needs conversion before numeric storage
  • Inline email images should be filtered before document extraction
  • Prebuilt models avoid training when a standard task already fits
  • Generic prediction models cannot be invoked from apps or flows
  • Custom models need training and publication before consumption
  • Publishing replaces the previously published model version
  • Model co-owners can retrain and publish but cannot delete

Unlock with Premium — includes all practice exams and the complete study guide.

Also tested in

References

  1. https://learn.microsoft.com/en-us/microsoft-copilot-studio/prompts-overview
  2. https://learn.microsoft.com/en-us/ai-builder/model-types
  3. https://learn.microsoft.com/en-us/credentials/certifications/resources/study-guides/ab-410
  4. https://learn.microsoft.com/en-us/ai-builder/use-a-custom-prompt-in-flow
  5. https://learn.microsoft.com/en-us/ai-builder/use-a-custom-prompt-in-app
  6. https://learn.microsoft.com/en-us/power-apps/maker/data-platform/prompt-column
  7. https://learn.microsoft.com/en-us/ai-builder/administer-licensing
  8. https://learn.microsoft.com/en-us/microsoft-copilot-studio/prompt-library
  9. https://learn.microsoft.com/en-us/microsoft-copilot-studio/create-custom-prompt
  10. https://learn.microsoft.com/en-us/microsoft-copilot-studio/add-inputs-prompt
  11. https://learn.microsoft.com/en-us/microsoft-copilot-studio/code-interpreter-for-prompts
  12. https://learn.microsoft.com/en-us/microsoft-copilot-studio/azure-openai-human-review
  13. https://learn.microsoft.com/en-us/microsoft-copilot-studio/use-your-own-prompt-data
  14. https://learn.microsoft.com/en-us/microsoft-copilot-studio/prompt-model-settings
  15. https://learn.microsoft.com/en-us/microsoft-copilot-studio/bring-your-own-model-prompts
  16. https://learn.microsoft.com/en-us/microsoft-copilot-studio/process-responses-json-output
  17. https://learn.microsoft.com/en-us/ai-builder/train-model
  18. https://learn.microsoft.com/en-us/ai-builder/manage-model
  19. https://learn.microsoft.com/en-us/ai-builder/publish-model
  20. https://learn.microsoft.com/en-us/ai-builder/form-processing-model-in-flow
  21. https://learn.microsoft.com/en-us/microsoft-copilot-studio/share-your-prompt
  22. https://learn.microsoft.com/en-us/ai-builder/share-model
  23. https://learn.microsoft.com/en-us/ai-builder/powerfx-in-powerapps
  24. https://learn.microsoft.com/en-us/ai-builder/use-in-powerapps-overview
  25. https://learn.microsoft.com/en-us/ai-builder/flow-text-recognition
  26. https://learn.microsoft.com/en-us/ai-builder/use-in-flow-overview
  27. https://learn.microsoft.com/en-us/troubleshoot/power-platform/ai-builder/power-automate/the-output-of-an-ai-builder-action-is-empty-in-power-automate-or-the-next-action-returns-an-error
  28. https://learn.microsoft.com/en-us/power-automate/guidance/coding-guidelines/asychronous-flow-pattern
  29. https://learn.microsoft.com/en-us/ai-builder/prebuilt-category-classification-pwr-automate
  30. https://learn.microsoft.com/en-us/ai-builder/flow-language-detection