AI CSV Cleaner: Set Rules Before Changing Your Data

Worked example

Preserve identifiers while normalizing dates

Illustrative example written for this guide. Model outputs vary.

Leading zeros can be meaningful. An ambiguous date should be reviewed, not silently guessed.

Prompt to adapt

Inspect this CSV before editing. Treat customer_id as text and preserve leading zeros. Trim outer whitespace in city. Identify the delimiter and report ambiguous dates. Propose duplicate rules and a change summary; do not delete rows merely because names match.

Inspect the file format first

Check the encoding, delimiter, header row and quoting. A comma inside a quoted address is part of the value, not necessarily a new column. Confirm that every row has the intended number of fields.

Set column types explicitly. Postal codes, telephone numbers and customer IDs are often text even when they contain digits. Keep a copy of the original export for comparison.

Define duplicates using the business key

Two rows with the same customer name may represent different people or transactions. Choose the fields that establish identity, and decide how to handle conflicting values before removing anything.

Ask for exact matches and possible matches in separate review groups. Preserve the retained record’s source and explain why another row would be removed.

Reconcile the cleaned export

Compare row counts before and after, count missing values by column and inspect a sample of changed cells. If rows were merged or removed, every difference should have an explanation.

Open the exported CSV in the system that will consume it. Verify identifiers, non-English characters and dates there; a correct preview alone does not prove a successful import.