Data Dictionary in Fabric Copilot: step by step

João Barros 18 de July de 2026 4 min read

The goal is to create a Data Dictionary in Fabric Copilot for a Lakehouse table: a clear description of the columns, examples of values and cleaning suggestions. Having a Data Dictionary improves data understanding, speeds up analysis and facilitates governance.

Prerequisites

Account with access to Microsoft Fabric and permission to view the Lakehouse where the table is located.
Table or file (CSV/Parquet) already uploaded to OneLake/Lakehouse.
Copilot active in your organization and access to the Copilot pane in the workspace.
A Notebook or area to save the result (for example, a Notebook in Fabric or a Markdown file in OneLake).

Step 1: Identify the table and gather context

Open the Fabric workspace and locate the table in the Lakehouse. Note the table name and, if possible, a small sample set (5–20 rows) so Copilot can infer example values. If you prefer, export 10 rows to a small CSV file.

Step 2: Open Copilot in the table context

With the table selected, open Copilot (chat/Copilot icon in Fabric). Ideally start Copilot with the table context (when Copilot has access to your workspace data, the model uses that information). If there is no automatic context, paste a small schema or sample into the chat.

Step 3: Ask Copilot to generate the Data Dictionary

Send a clear prompt to Copilot, requesting a Data Dictionary with the columns and fields you want. Use this example prompt adapted to your case:

Cria um Data Dictionary para a tabela 'sales_orders'. Para cada coluna, fornece: tipo provável, descrição concisa, exemplo de valor, percentagem de nulos (estimada), e 1-2 sugestões de limpeza. Mostra o resultado em formato de tabela Markdown.

Example prompt with data sample (when there is no direct access to the table):

Aqui estão 5 linhas de 'sales_orders':
OrderID,CustomerID,OrderDate,Amount
1001,C123,2024-01-10,250.00
1002,C456,2024-01-12,
1003,C123,2024-01-15,75.50
1004,,2024-01-20,300
1005,C789,2024-01-20,120.25
Com base nisto, cria o Data Dictionary como pedido acima.

Step 4: Review and refine the result

Copilot will return a Markdown table with columns like Column Name, Likely Type, Description, Example, % Nulls, Suggestions. Check for unexpected values and ask for refinements: for example, request separating integer/decimal numeric types, or suggest concrete ETL rules.

Example output (summary):
| Coluna | Tipo provável | Descrição | Exemplo | % Nulos (estim.) | Sugestões |
|---|---|---|---:|---:|---|
| OrderID | inteiro | Identificador único do pedido | 1001 | 0% | Verificar duplicados, chave primária |
| CustomerID | string | Identificador do cliente | C123 | 10% | Substituir vazios por 'Unknown', validar formato |
| OrderDate | date | Data do pedido | 2024-01-10 | 0% | Normalizar formato ISO; validar intervalos |
| Amount | decimal | Valor total do pedido | 250.00 | 20% | Tratar nulos, converter para decimal com 2 casas |

If Copilot is wrong about types or examples, explain why you think it is incorrect and ask for a new version. For example: "OrderDate appears to contain timestamps — adjust the type to datetime."

Step 5: Save and integrate the Data Dictionary

Once satisfied, copy the Markdown table to a Notebook in Fabric (create a Markdown block) or save as a .md file in OneLake. To integrate with governance, paste shorter descriptions into the table's column properties (description field) in the Fabric data catalog or in the metadata manager your organization uses.

# Quick steps to save:
1. New Notebook → New Markdown block → Paste Markdown table
2. Save Notebook in the workspace/OneLake
3. (Optional) Update column descriptions manually in the Fabric catalog

Verify the result

Confirm the Data Dictionary is complete: each column has a description, an example and at least one cleaning suggestion. Open the Notebook or .md file and review the table. Test 1–2 cleaning suggestions (for example replacing nulls) on a sample to validate the recommendations. If Copilot generated estimated % nulls, calculate them with a query or sample to confirm.

Conclusion

Having a Data Dictionary generated with Copilot in Fabric speeds up exploration and documentation of your data. Next steps: automate Data Dictionary generation for new tables (manual workflow with standard prompts) and integrate metadata into the catalog. Tip: create a standardized prompt for your team to ensure consistency — do you already have a preferred prompt to generate descriptions?