Writing Good SQL Prompts

AI assistants like ChatGPT, Claude, Gemini or GitHub Copilot can write, fix and explain SQL in seconds — but only as well as you ask. A vague question gets a generic guess. A precise question gets a query that runs on your tables. This chapter gives you 16 copy-paste prompts that work, all tried on our IPL data.

Definition

A prompt is the instruction you give an AI assistant. A good SQL prompt tells the AI what to do (task), what it is working with (context: database engine, tables, columns), and what the answer must look like (specifications).

The AI cannot see your database. Everything it knows about your tables comes from the prompt. Every missing detail becomes a guess, and guesses produce wrong column names, the wrong SQL dialect, or a query that answers a different question.

The five prompt components

Tip: Always name the database engine. TOP 3 works in SQL Server, but MySQL and PostgreSQL need LIMIT 3. LEN() is SQL Server, LENGTH() is MySQL/PostgreSQL. If you don't say which one you use, the AI picks one for you.

Weak prompt vs strong prompt

Weak:

Write SQL for players

• ❌ Which database? ❌ Which tables and columns? ❌ What should the query return?
• Result: a generic guess that probably won't run on your data.

Strong:

Act as a senior SQL developer.                                   ← Role
In my SQL Server database I have:
players (id, name, team, country, runs).                        ← Context
Write a query that ranks players by runs.                        ← Task
Return id, name, team, runs, rank. Add comments.                 ← Specifications
Explain your solution simply.                                    ← Tone

The more detail you put in the prompt, the fewer wrong guesses the AI makes.

Where AI can help you with SQL

There are two kinds of assistants, and they are good at different jobs:

The "Do the following:" structure

Every prompt in this chapter uses the same skeleton. It works because the AI answers each numbered point in order, so nothing gets skipped:

<one or two lines of context: engine, tables, the situation>
Do the following:
1. <first thing>
2. <second thing>
3. <third thing>
[ SQL Query GOES HERE ]

Common mistakes

• ❌ Leaving out the table structure, so the AI invents columns like player_name or total_runs. ✅ Paste the column list or the CREATE TABLE.
• ❌ Asking for "the best query" with no definition of best. ✅ Say "best for readability" or "fastest on a large table".
• ❌ Accepting the first answer. ✅ Iterate: paste the error or a wrong result back and ask again.