5.2 Configuring Prompt Columns and Row Summaries

Prompt Columns

A prompt column is an AI-powered data type in Microsoft Dataverse: instead of a user typing a value, the column's content is generated by a natural-language AI prompt that you ground in other columns of the same row. The generated result is stored persistently in the table, so it behaves like ordinary data — available to views, forms, workflows, and reports.

Typical uses include condensing a long customer feedback column into a two-sentence brief, classifying sentiment, extracting action items from notes, or drafting a reply suggestion from a support inquiry.

Creating a prompt column

1. In make.powerapps.com, open Tables, open your table, and select New → Column.
2. Give the column a display name, then set Data type to Prompt.
3. Select +Add new prompt and write the instruction. Ground it with +Add content, choosing one or more input columns from the table.
4. Save the prompt, then save the column.

Key limits and rules:

• A table supports up to five prompt columns.
• Input columns cannot be formula columns, file columns, image columns, or another prompt column — those inputs are ignored.
• A default text input variable, applied as a filter on the table's primary column, is required in the prompt definition; saving fails without it.
• The environment's Block unmanaged customizations setting must be off to create or edit prompt columns.

Testing before you ship

Create a dedicated test record with realistic values in every input column. While editing the prompt, select the input column to open Filter knowledge, choose Filter attribute, and point it at your test record. Select Test to review the Model response, and open the Knowledge used tab to confirm the prompt read the record you intended. Iterate on the wording, set the filter back to No filter, save — and finish by checking real output on a model-driven app form or view.

When values are generated — and when they are not

Prompt columns compute asynchronously, decoupled from the save transaction so record operations stay fast. The execution triggers are precise, and the exam probes them:

Every prompt column automatically gets two companion columns, (name)_PromptColumnStatus and (name)_PromptColumnDetails. Status codes: 0 NotStarted, 1 InProgress, 2 Completed, 2000 (skipped — filter conditions not met), 2001 (skipped — no referenced input column changed), 3 Failed (error text in the Details column).

You can narrow execution with filter conditions so the prompt runs only when a condition is true (for example, only when Status equals Resolved) — this conserves Copilot credits. Generation can be switched off per column with Allow prompt column execution, and the feature also requires the tenant/environment-level Copilot and AI Prompts feature settings; both levels must be enabled for execution to happen.

Licensing and governance (awareness level)

• Prompt columns use the same licensing model as AI Builder prompts — execution consumes AI Builder / Copilot credits, and missing entitlements or exhausted credits are a documented cause of failed generation.
• Prompt columns are not audited, so do not rely on the audit log to reconstruct who generated what.
• A common failure cause is a user lacking read permission on an input column (column-level security applies to the prompt's grounding data), which produces status 3.

Exam trap: candidates often assume prompt columns recalculate on a schedule or when the record is opened. They do neither — the only refresh paths are record creation and updates to the columns the prompt references.

Row Summaries

A row summary is a Copilot-generated, natural-language digest of a Dataverse row, surfaced in model-driven apps either as a collapsible summary bar at the top of a main form or per row from views. It exists to save users from scanning dozens of columns spread across tabs and related tables.

Enabling and configuring

1. Prerequisite (admin): a Power Platform administrator enables the AI insight cards environment setting (Power Platform admin center → environment → Settings → Features). The setting defaults to On, and newer admin surfaces also expose it under Copilot → Settings → Power Apps → Summary.
2. Maker configuration: in make.powerapps.com, open the table and under Customizations select Row summary. If the option is disabled, hover over it to see why — for example, the table must contain at least one row of data.
3. In the Prompt box, add the columns to summarize by selecting Add or typing /. You can steer formatting (bulleted list versus paragraph) and request generation in the user's preferred language.
4. Select Test prompt — the preview is generated from the most recently edited row in the table.
5. Select Apply to main forms.

Scope rules the exam cares about:

• The configuration is per table and applies to all views and main forms of that table, in every model-driven app that uses them. Forms carrying a summary show a form AI icon in the table's Data experiences → Forms list.
• Tables that already ship their own app summaries — Case, Lead, and Opportunity (Dynamics 365 Customer Service and Sales) — are excluded from this feature to avoid conflicts.

Prompt columns vs row summaries

Choose a prompt column when downstream logic (flows, reports, business rules) needs the AI-generated text as data; choose a row summary when the goal is a quick, human-readable orientation at the top of a form.