7 Spreadsheet Problems That Give AI Tools Wrong Answers

AI tools are very good at reading tables and very bad at knowing when a table is misleading. They do not stop to ask whether a row is a total, or whether two customer names are the same business. They take the data as it is and answer.

Here are the seven problems we see most often in small business spreadsheets, what goes wrong, and how to fix each one.

1. Totals rows inside the data

A "Total" row at the bottom of a list, or subtotals after each month, get counted as if they were records. Ask for total sales and the answer can come back doubled.

Fix: keep totals out of the data. Put them above the table or on a separate sheet, or let a PivotTable or dashboard calculate them.

2. Titles above the header row

A report title, date or logo in the first few rows means the real column names start on row 4 or 5. Tools can mistake the title for the headers and misread every column below it.

Fix: start the table on row 1 with a single header row. Move titles and notes to another sheet.

3. Blank or duplicate column headers

Two columns called "Amount", or a column with no name at all, leave the AI guessing which one you mean.

Fix: give every column a unique, descriptive name, such as "Amount ex VAT" and "Amount inc VAT".

4. Duplicate rows and duplicate IDs

The same invoice pasted twice, or two customers sharing an ID, inflates counts and totals. The AI has no way of knowing that one of them should not be there.

Fix: run Data, Remove Duplicates on a copy of the file, and check that ID columns hold unique values.

5. Numbers stored as text

Values that look like numbers but are stored as text can be skipped or read differently. This is one of the most common reasons totals do not match.

Fix: convert them back to real numbers.

6. The same thing spelled several ways

"Glasgow", "glasgow" and "Glasgow " with a trailing space can be treated as three different places. Ask which region sold most and the answer is split across them.

Fix: standardise labels with Find and Replace or TRIM, then add a dropdown list so new entries stay consistent.

7. Mixed date formats

A column holding 04/05/2026, 2026-05-04 and "4 May" is a column of guesses. UK and US date order makes it worse: 04/05 can mean April or May.

Fix: convert the column to real Excel dates and apply one format. If the data comes from an export, set the date type in Power Query.

Find them in under a minute

You can check for all seven by hand, but it takes time on a large file. The free Data Health Check looks for every problem on this list and gives your spreadsheet a score out of 100.