As of 13 August 2026, AI can only partly write SQL queries for your business database.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open your database documentation, administration console or schema export and gather the SQL dialect, database version, relevant table names, columns, data types, primary keys and foreign keys.
   2. Remove customer names, email addresses, account numbers and other unnecessary personal data, then paste the remaining schema and the exact reporting question into the prompt.
   3. Add the required date range, filters, grouping rules, exclusions, sorting and expected output columns, defining any business term that could have more than one meaning.
   4. Add anonymised test cases or known totals, select read-only access in the database tool, and ask the model to produce the query and validation checks using the prompt.
   5. Run the query first in a development environment or against a limited sample, then compare its row counts, totals and sample records with your known test cases.
   6. Inspect the execution plan or ask your database administrator to check joins, indexes, permissions, personal-data exposure and load before running it on the live database.
   7. Save the verified query with its schema version, assumptions and owner, then send it to the report or automation that needs the result.

   Prompt

   Write a SQL query for the business reporting task below. Use only the supplied schema and facts. Do not invent tables, columns, relationships, values or business rules. If anything needed is missing, ask focused questions before writing the query.
   
   SQL dialect and database version: [for example, PostgreSQL 15, MySQL 8, SQL Server, or SQLite]
   
   Database schema:
   [Paste CREATE TABLE statements, a schema export, or a complete list of tables, columns, data types, primary keys and foreign keys. Do not paste customer names, contact details or other unnecessary personal data.]
   
   Reporting task:
   [Describe exactly what the result must show, including filters, date range, grouping, sorting, inclusion and exclusion rules, and the meaning of any business terms.]
   
   Expected output columns:
   [List the columns and labels you want returned.]
   
   Known checks:
   [Give one or more records, totals, or test cases whose correct result is already known, using anonymised or synthetic values where possible.]
   
   Requirements:
   1. Produce a read-only SELECT query only. Do not write INSERT, UPDATE, DELETE, DROP, ALTER or other data-changing statements.
   2. Explain each join, filter, grouping choice and assumption in plain English.
   3. Flag any ambiguity instead of choosing silently.
   4. Warn about duplicate rows, NULL values, date and time-zone handling, and performance risks where relevant.
   5. Provide a small set of validation queries or checks that I can run against known results.
   6. Format the final SQL clearly and identify the SQL dialect it uses.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.