Expert guide
Ask Your Access Database Questions in Plain English
How natural-language queries work, when to trust the answer, and how to set your database up for it
Call us at +1 (385) 386-3860
Short answer: a natural-language query tool reads your Access tables and relationships, uses an AI model to turn a plain-English question into a SQL SELECT, runs it read-only, and shows the answer with the SQL it used. Accuracy depends on clear field names, defined relationships and unambiguous questions. Choose a tool that blocks every write query and shows its SQL. DatabaseAI does this for Access files on Windows.
In most companies that run on Access, there is one person who can write queries. Everyone else asks them: how many open jobs does the North team have, which suppliers have we not paid in 60 days, what did we sell to this customer last year? Each question takes that person ten minutes and an interruption. Natural-language query moves those questions to the people who ask them.
How Natural Language Becomes SQL
1. Read the schema
The tool lists tables, fields, data types and relationships in the Access file.
2. Interpret the question
The question and the relevant schema are sent to an AI model, which writes an Access SQL SELECT statement.
3. Validate
The SQL is checked: it must be a SELECT, reference real tables and fields, and contain no action keywords.
4. Run read-only
The query runs on a read-only connection to the database.
5. Present and explain
The result appears as a sentence, a table or a chart, with the SQL available for checking.
Questions It Handles Well
| Question | What the SQL does |
|---|---|
| How many open work orders does each technician have? | GROUP BY technician with a status filter |
| Which customers ordered over $10,000 this year but nothing last month? | Two date-range aggregates combined |
| List invoices more than 60 days overdue with the customer phone number | Join invoices to customers, date arithmetic on due date |
| What were monthly sales by product category in 2026? | Aggregate by Format(date, 'yyyy-mm') and category |
| Which items are below their reorder level? | Compare on-hand quantity to a reorder field |
Where Answers Go Wrong, and How to Prevent It
Ambiguous field names
If tblOrders has Date, Date2 and DT fields, no human or AI can tell which is the order date. Renaming fields to OrderDate, ShipDate and InvoiceDate is the single biggest accuracy improvement. If renaming would break forms, add field descriptions in table design; good tools read them.
Undocumented codes
A Status field with values 1, 2, 3 and 9 means nothing without a lookup table. Create small lookup tables (tblStatus: 1 = Open, 2 = Scheduled ...) and relate them. AI then writes "WHERE Status = 'Open'" correctly through the join.
Missing relationships
Access lets you join tables in queries without defining relationships. Define them in the Relationships window so the tool knows how customers, orders and lines connect. As a bonus, enforced referential integrity stops orphan records.
Ambiguous business terms
"Sales last month" might mean orders placed, invoices raised or cash received. Agree definitions and write them into a short glossary for the team, then phrase questions precisely: "invoiced amount in October 2026".
Safety Checklist for Any Tool
Opens the database read-only, not just 'promises' not to write.
Blocks UPDATE, DELETE, INSERT, make-table and DDL statements.
Shows the SQL for every answer.
Is clear about what leaves the computer: schema and question are needed; record data should not be.
Respects that some tables (payroll, HR) may need to be excluded.
Works on a copy or front end without locking the back end for other users.
DatabaseAI is a Windows desktop app that opens .accdb and .mdb files read-only, runs only SELECT queries, and keeps rows on your PC; details and pricing are on the AI for Microsoft Access page.
DIY with ChatGPT vs a Dedicated Tool
| Aspect | Copy schema into ChatGPT | Dedicated NL query tool |
|---|---|---|
| Who can use it | Someone who can paste SQL into query design | Any staff member |
| Schema handling | Typed or pasted each time | Read automatically |
| Running the query | Manual | Automatic, read-only |
| Write protection | Depends on the person | Enforced |
| Charts and export | Manual | Built in |
| Cost | Assistant subscription | Per-user licence |
The DIY route is covered in using ChatGPT to write Access SQL.
Preparing Your Database in an Afternoon
Rename the ten most-used ambiguous fields, or add descriptions.
Add lookup tables for coded fields.
Define relationships between the main tables.
Write a one-page glossary of business terms.
List ten real questions staff asked last month and test them.
If your database needs more than an afternoon, such as normalising duplicated tables or fixing years of inconsistent data, an Access consultant can restructure it without breaking existing forms. For recurring outputs, keep using proper Access reports; for whether Copilot can do any of this, see does Copilot work with Microsoft Access.
FAQ
Natural-language Access queries FAQ
Accuracy, safety and setup