Clean a Spreadsheet Before Asking AI to Analyse It
A spreadsheet analysis can look authoritative while resting on inconsistent dates, duplicate rows or mixed units. AI can help identify suspicious patterns, but the source data needs inspection first. Establish what one row represents and what each field means before asking for conclusions or charts.
Why it matters
Cleaning decisions change results. Removing a repeated row may be correct for a customer list and wrong for a transaction log. The dataset's meaning matters more than making every cell look uniform.
A practical workflow
- Keep an untouched copy and document what each row, column, unit and missing value represents.
- Check identifiers, repeated records, date formats and numeric fields before grouping or calculating averages.
- Use AI to suggest questions or candidate inconsistencies, then verify each proposed correction against the source.
- Record cleaning rules and compare totals before and after changes so unexpected losses remain visible.
An example
An attendance sheet may contain two entries for the same person because they attended two sessions. Deduplicating only by name would erase a valid record. Define whether the analysis concerns unique people, registrations or session visits before cleaning the data.
A mistake to avoid
Do not upload personal data to an unapproved tool or accept an unexplained correction in bulk. A model may infer missing values that look reasonable but were never recorded. Keep inferred data separate from observed data.
Try it and check the result
Manually calculate a small sample and compare it with the analysis. If the totals differ, trace the discrepancy through the cleaning rules before creating charts or writing a conclusion from the full dataset.
Reference and further reading
Google's prompt design strategies.
The workflows and examples in this guide are illustrative. Product controls and availability can change; check the linked official guidance for your setup.
Continue with Run a Weekly Review with AI Without Creating More Admin or explore all blog topics.