Check a money spreadsheet for broken totals

A cell-by-cell review list of suspected total-formula errors with proposed corrections, applying no live edits.

Choose one of three approaches. The steps for that option appear below.

Choose an option above. On a phone, swipe the tabs left or right to see all options →

ChatGPT: lists suspected formula errors, you check

ChatGPT checks each total formula against your row map and proposes corrections. You check every flag against the real sheet.

  1. You

    Copy the extract

  2. ChatGPT

    List suspected errors

  3. You

    Check flags and save

  1. Copy the formulas into an extract

    Use: You

    Have ready: the spreadsheet. Copy each total cell's formula and ranges into an extract; map line items, subtotals and intended rows (not stated if unclear), plus referenced helper formulas for tracing double counting.

    Check

    Copy each formula character for character. Keep row numbers accurate; the whole point is which rows are in each range.

  2. List suspected errors in each formula

    Use: ChatGPT

    Ask ChatGPT to check every total formula against the row map and list each suspected skipped row or doubled cell with a proposed correction.

    Details

    What to paste into the tool

    Task: Review the spreadsheet formulas in the extract below. For each total or summary cell: quote its formula, state the range it covers, and check that range against the row map and my explicit intended rows. Trace nested subtotal/helper references if supplied; do not infer business intent or assume a row belongs in a total. When intention or helper formulas are missing, mark the check 'not stated'. Flag any range that appears to skip a row that should be included, include a row that should not be there, or add a cell twice. For each flag, propose the corrected formula for my review; do not present any correction as final. Check only formulas and ranges; no comments on the money. Write 'not stated' where the extract does not show something you need. Inputs: [upload the extract file, or paste the extract with formulas and the row map into the chat]. Task: Turn my confirmed review decisions into a tidy cell-by-cell list. For every reviewed total cell, including those with no suspected fault: the current formula/range, my decision or no issue identified, and any correction I accepted, marked 'for me to apply'. Apply only my decisions; do not add new flags or changes. Inputs: [upload or paste your confirmed decisions list].

  3. Check each flag and save the list

    Use: You

    Check each suspected error and proposed correction against the real sheet; ask for a tidy final list in the same chat, then recheck and save.

    Check

    Open the real sheet and look at each flagged cell; a flag is suspicion, not fact, a proposal a draft, not a fix. Check each proposed formula covers the intended rows. Any later live edit and resulting-total check happen personally, outside this review workflow. The assistant never touches the live sheet.

Before you paste into any AI tool, remove any confidential information. Put it back at the end.

Tools in this workflow

Deciding between ChatGPT and Claude for this? See the comparison.

Final check

Never paste credentials or full payment details.

Check the formula notation in your spreadsheet software; mark missing or unchecked details not stated. Steps marked Human are yours to do. Human with an assistant means you give the assistant the stated inputs and check its output before using it.

Nothing is sent, signed, paid or published by an assistant.

A human check follows every assistant step and the last step is human.

Check the result against your original inputs. Approve it before sharing or acting.

Availability notes

Tool pages were read signed out on 2026-10-28. Plan, price and free-limit details were not re-checked and are not claimed.

What still needs checking
  • Whether any assistant named here does the arithmetic or rule-following in this scenario correctly: no page evidence, so you check cover it.
  • Which plan each quoted capability needs.
  • You decide any open choices named in the scenario itself; none were guessed.

Share this page