Ask for SQL you can check

A prompt is the message you send to the AI. A schema is the list of your tables and their columns. Without them, the AI invents plausible names and picks its own definitions.

A SQL prompt you can check has four parts:

• Tables: names, key columns and how they join.
• Definitions: what "revenue" or "active customer" means at your company.
• Rules: the date range, the time zone, and what to leave out, like refunds or test accounts.
• Return: the query, every assumption it made, and a second query that counts the rows.

The assumptions list is the part you'll use most. It shows you exactly where the model guessed.

A worked example

Write a PostgreSQL query for monthly revenue by country.

Tables: orders(order_id, customer_id, amount_usd, status, created_at),
customers(customer_id, country). Join on customer_id.

Definitions: revenue = sum of amount_usd where status = 'paid'.

Rules: created_at is in UTC. Use 1 Jan to 30 Sep 2026.
Leave out customers whose email ends in @test.example.

Return: the query, a list of every assumption you made, and a query
that counts the orders included so I can check the result.

Your turn:

Write a [dialect] query for [question].
Tables: [names, key columns, how they join]
Definitions: [metric] = [exact rule]
Rules: [date range], [time zone], [what to leave out]
Return: the query, every assumption, and a row-count query.

Try it