From a question to a query you can check.
Describe the data you need. The prompt carries your schema, names the dialect, and asks for the query with an explanation, so you can verify it before you run it.
What the prompt you get back gets right
Schema in, guesswork out
Your tables and columns go into the prompt, so the model works from your actual names instead of inventing plausible ones.
Dialect named
Postgres, MySQL, BigQuery and the rest differ in the details — the prompt states which one you are on.
Query plus explanation
You get the SQL and a plain-language walkthrough of what it does, in that order.
Guardrails included
Read-only, no destructive statements, a row limit, and every assumption about your data listed explicitly.
From a rough note to a working prompt
Your note
“which customers ordered twice last month but haven't since?”
The prompt you get back
- Role
- an analyst who works only from the schema you provide
- Schema
- the tables and columns involved, pasted in as-is
- Task
- your question, stated as the result you want back
- Output
- the query first, then a plain-English explanation, then any assumptions it made
- Guardrails
- read-only, no destructive statements, row limit, and flagged assumptions
Every prompt is written for your exact request — this is the structure it comes back in, not a fixed template.
Common questions
Is it safe to run what comes back?
The prompt asks for a read-only query plus an explanation and its assumptions. Read the explanation, check the assumptions against your data, and run it with a row limit or against a copy first.
What if my schema is large?
Paste only the tables involved. The prompt requires the model to state any assumption it makes about columns it was not given, rather than silently guessing.
Try it with your own notes
One pass in, a structured prompt out — saved to your library so you can reuse it.
Engineer my promptOther tasks people engineer prompts for: