Pre-prompts are how your organization teaches Copilot its standards once so every team member gets consistent behavior. This article covers the Copilot settings screen, how to write an effective pre-prompt, and how the Copilot role picker differs from the platform license role.
Who this is for: Project administrators, team leads, and architects responsible for standards and conventions. Pre-prompts are configured at the project level and apply to every user working in that project.
What pre-prompts do
A pre-prompt is a set of standing instructions attached to the project. Every time anyone on the team runs a Copilot prompt, whether it’s reverse engineering, a question about the model, or a bulk governance action, the pre-prompt is sent to the AI along with the user’s prompt. That means:
- You don’t have to repeat your conventions in every prompt.
- Team members get consistent output regardless of prompt-writing skill.
- New team members inherit the standards immediately without onboarding.
Good pre-prompts describe how your team works: architecture patterns you follow, naming conventions, expected data types, governance rules, and the shared vocabulary your business uses.
Opening the Copilot settings screen
- Open the project you want to configure.
- In the Copilot panel on the left-hand side, click the settings icon (gear).
- The settings screen opens with two tabs: Pre-prompt and Role.
Only users with project admin rights can edit Copilot settings. Other users can view the settings but cannot change them.
Account administrators and alternative administrators can also edit Copilot settings at the account level.
Writing an effective pre-prompt
The best pre-prompts are specific about what the team does and does not do. Vague pre-prompts (“follow best practices”) don’t give Copilot enough to work with. Here’s a realistic example for a team using the Medallion architecture:
Example pre-prompt: Our team follows the Medallion architecture with three layers: bronze (raw source data), silver (cleansed and conformed), and gold (dimensional models for reporting). All fact tables are prefixed FCT_, all dimensions are DIM_, and all staging tables are STG_. Date columns use _DATE suffix and datetime columns use _TS. Primary keys are named <table>_ID. Before creating a new entity, check whether a similar dimension already exists in the project and propose reuse instead of duplication. Flag any column that looks like PII with the PII tag.
Write your pre-prompt in the same tone you’d use for a new engineer joining the team. State the architecture, conventions, and what to check before creating new objects. Update the pre-prompt over time as your standards evolve.
The Copilot role picker
Copilot responses are calibrated to the role you’re working as. Roles available today:
Modeler / Architect
You design schemas, standards, and structures. AI helps you generate data models, enforce naming conventions, run impact analysis, and produce documentation that stays current as your designs evolve.
BI / Documentation Lead
You make data models readable by everyone. AI helps you auto-fill missing descriptions, flag undocumented objects, and generate documentation a business analyst can actually use.
Data Governance
You own the rules, definitions, and compliance posture. AI helps you scan for PII and policy violations, review change proposals against governance standards, and maintain an audit-ready record of what changed and why.
Consumer / Viewer
You use data to answer business questions. AI helps you understand what tables and fields mean, explore the data model without needing a data engineer, and find the right data asset for what you are trying to do.
Analytics Engineer
You sit between raw data and the business. AI helps you generate semantic layer definitions, produce dbt YAML, and keep your analytical layer in sync as the underlying schema evolves.
Other / General
You work across SqlDBM in ways that do not fit a single role.
The role setting is per-user and applies across every project you work in. You can change it any time from Copilot settings.
Important: The Copilot role is separate from your SqlDBM license role (modeler, reviewer, consumer, etc.). The license role controls what you can do in the platform: create objects, edit descriptions, read-only access. The Copilot role only controls how the AI phrases its responses. You cannot grant yourself extra permissions by changing your Copilot role.
When pre-prompts don’t help
Pre-prompts are strong at conventions and context, weak at facts the AI has no other way to know. If your business has a definition of “active customer” that depends on a SQL query against live data, Copilot can’t enforce it from metadata alone. Pre-prompts work best for:
- Architecture patterns (Medallion, Data Vault, star schema).
- Naming conventions and prefixes.
- Data type preferences (e.g., all amounts use DECIMAL(18,4)).
- Rules about object creation (“check before duplicating,” “always create audit columns”).
- Terminology guidance (“when we say ‘client’ we mean the legal entity, not the user”).
Iterating on your pre-prompt
Your first pre-prompt will not be perfect. Expect to refine it over time as you watch how Copilot behaves with your team. A practical pattern:
- Start with a short pre-prompt covering your top 3–5 rules.
- Ask the team to run real prompts for a few days and flag where Copilot output doesn’t match expectations.
- Update the pre-prompt to address the specific gaps.
- Review the pre-prompt quarterly; prune rules that are no longer relevant.
Related articles
- Getting started with Copilot
- Reverse engineering a model with Copilot
- Bulk governance actions
- Enabling Copilot for your account and users
- Settings and Conventions