How to create data translation prompts in Fabric Copilot

João Barros 15 de August de 2026 3 min read

This tutorial shows how to use Copilot in Fabric to translate column descriptions and table metadata (for example, from English to European Portuguese), preserving technical terms and abbreviations intact. Translating metadata with context preservation is useful for multicultural teams and for creating consistent documentation in Power BI or Warehouse.

Prerequisites

Account with access to Microsoft Fabric and permissions to edit Dataflows or Tables in the Lakehouse/Warehouse.
A table with columns and descriptions (metadata) already existing in English.
Copilot enabled in your Fabric workspace.

Step 1: Prepare a representative example

Choose a small table (5-10 columns) with descriptions that contain technical terms, units and abbreviations. Creating a set of examples helps Copilot understand the pattern and keep abbreviations.

# Exemplo de estrutura (não é código Fabric, é só referência)
ColumnName | Description
OrderID    | Unique identifier for order
ShipDate   | Date order shipped (UTC)
AmtUSD     | Order amount in USD
CustID     | Customer identifier (FK)
Status     | Current order status: Placed, Shipped

Step 2: Open Copilot in the metadata context

In the workspace where the table resides, open the item (Dataflow/Table) and locate the metadata/descriptions pane. Click Copilot to open the conversation window in the same context — this allows Copilot to access the current descriptions and the schema.

Step 3: Craft the initial prompt — clear instructions

Explain exactly what you want: target language, terms to preserve (for example, UTC, FK, abbreviations like ID, USD) and the tone (formal/informal). A clear prompt reduces revisions.

Translate the following column descriptions from English to European Portuguese. Preserve technical terms and abbreviations (UTC, ID, FK, USD). Keep units and examples intact. Provide only the translated descriptions in a JSON array keyed by ColumnName.

[PASTE CURRENT DESCRIPTIONS HERE]

Step 4: Adjust with preservation examples

If Copilot translates abbreviations you want to preserve, give an explicit example in the prompt. Include 1–2 example lines showing what should remain unchanged.

Example: "OrderID: Unique identifier for order" -> "OrderID: Identificador único da encomenda" (preserve "OrderID" and "ID")

Now translate the real descriptions.

Step 5: Review and request variations

Read the generated translation and look for common errors: translation of abbreviations, units changed, technical terms improperly translated (for example, 'Warehouse' vs 'armazém' if you prefer to keep the product name). If necessary, ask Copilot for a shorter variant or a different tone.

Can you shorten each description to a maximum of 80 characters keeping the technical terms intact?

Step 6: Apply the translated descriptions

Copy the generated JSON or list and paste it back into the table metadata pane in Fabric. For Dataflow, update the 'Description' fields; for Tables in the Lakehouse/Warehouse, edit the column properties.

Verify the result

Confirm that each column has the correct translated description, that abbreviations are preserved and that no inconsistencies were introduced (for example, changing UTC to local time). Test in Power BI by reimporting the table or viewing the schema to ensure the descriptions appear as expected.

Conclusion

Translating metadata with Copilot in Fabric speeds up localization of models and collaboration between teams. Next steps: create standardized prompts for other tables or automate with a script that applies translations in batch via API (where applicable). Tip: keep a glossary of terms to preserve and always review technical translations before production.