How assistants read tables and spreadsheets on screen

TL;DR:

• A table is meaningful only if you know which header belongs to which column. Merged cells and hidden rows blur that link.

• An assistant reading a screen sees what is visible; one reading the file sees the cells, including hidden ones. The two can differ.

• Check totals against the visible numbers before you trust a summary.

You share your screen with an assistant and ask, "How many pens did we sell in March?" On the sheet, you see a block of numbers and you point at a column. The assistant answers 55. Is that right? It depends entirely on whether it matched the number to the correct row and the correct column header, and spreadsheets are good at hiding that connection.

The short answer

A table is a grid where position carries meaning. The number 55 means nothing alone. It means "pens, March" because of the row label to its left and the header above it. A person reading a sheet follows those links without thinking. An assistant has to rebuild them, from either the pixels it sees on screen or the cell data in the file, and each source has its own ways of going wrong.

What happens step by step

1. Find the table. It locates the grid on the screen or in the file, separating it from titles, notes and side panels.

2. Find the headers. It decides which row or column holds labels instead of data.

3. Map every number to a pair of labels. A row label and a column label per cell.

4. Handle special cases, such as merged cells, blank cells, subtotals and notes.

5. Answer from the mapping and say which cell it used.

Reading from a screenshot adds one more step at the start: working out which characters are in the picture. A small font or a blurry capture can turn a 3 into an 8. Reading the underlying file avoids that, but sees things you do not, such as hidden rows.

Walk-through: a sheet with merged cells

Here is a small, invented sales sheet. The words "Quarter 1" sit in one merged cell that spans three month columns.

The danger is the merged cell. In the file, "Quarter 1" lives in only one cell, the leftmost; the other two cells under the merge are empty. A reader who looks at columns one at a time may find that Feb and Mar have no group label, even though a person sees that they belong to Quarter 1. So a careful assistant treats the merge as covering all three.

Now the question: pens sold in March. Go to the row "Pens", then the column "Mar": the answer is 55. Next a harder one, the Quarter 1 total for both items.

1. Notebooks: 40 + 35 + 45 = 120.

2. Pens: 60 + 50 + 55 = 165.

3. Total: 120 + 165 = 285.

4. Cross check by month: Jan 40 + 60 = 100, Feb 35 + 50 = 85, Mar 45 + 55 = 100, and 100 + 85 + 100 = 285. Both routes agree.

Hidden data

Suppose a third row, Staplers, is hidden: 10, 12 and 8 for the three months. A screen shows only Notebooks and Pens, so a reader working from the screen would say the quarter total is 285. But a total cell written as a SUM over all three rows includes hidden rows and shows:

Staplers: 10 + 12 + 8 = 30
Quarter total including hidden row: 285 + 30 = 315

The sheet says 315 while the visible rows add to 285. A mismatch like this is a clue: the gap of 30 equals the Staplers row exactly, 315 − 285 = 30. A formula that skips hidden rows, such as SUBTOTAL with the option that ignores them, would give 285. Filtered rows behave similarly, so know which kind of total the cell is.

Reconcile first: When a total on the sheet differs from the sum of what you can see, suspect hidden or filtered rows before suspecting the arithmetic.

What it can't do

• See what is not shown. From a screenshot, hidden rows, other sheets and cell formulas are invisible.

• Always read tiny text. Cells with long numbers or small fonts can be misread.

• Guess missing headers. If a column has no label, it can only describe the numbers, not what they mean.

• Know units. A column of 45 might be dollars, units or minutes unless something says so.

Privacy and control

Spreadsheets often hold the most sensitive material in an office: salaries, customer lists, account numbers. Before you share a sheet, close tabs you do not need, hide nothing you are trying to protect (hidden is not deleted) and remove columns the question does not need. On screen, you can show a tool one block rather than the whole file. The habits in data safety basics for AI tools at work apply here.

A tool that looks at your screen only when you ask, as DuckyHelper does, limits when this can happen. It looks while the pill is open and for that answer.

Try it yourself

Build the sample sheet above in any spreadsheet. Merge the "Quarter 1" cells, hide the Staplers row, and ask your assistant for the quarter total. If it says 285, it read the screen. If it says 315, it read the cells. Then ask which rows it used. Knowing which kind of reader you have tells you how much to trust it.

Frequently asked questions

Why do merged cells cause trouble?

Only one cell holds the label; the others are empty. A reader going column by column can miss that the label covers all of them.

Is a screenshot or the file better?

The file has more detail, including hidden data. The screenshot shows exactly what you see. Each can mislead in its own way, so check totals.

How can I tell if rows are hidden?

Look for a jump in row numbers, such as 2 then 4. A hidden row leaves a gap.