How to get the data out of a spreadsheet that only appears on screen

Last updated 2026-09-17 · 8 min read

What is the short version?

A table on screen is a picture of data rather than the data itself. Reading it back from frames means optical recognition of small digits through video compression, and a single misread character inside a figure is considerably worse than having no figure at all.

1. Ask Vizard Agent to find the source address in the frame.
2. If it is reachable, fetch the real data instead of reading pixels.
3. Ask for the narration alongside it, so the rows have context.

What do you need before you start?

The video, and an idea of what the document is for. Vizard Agent produces a written document, a spreadsheet, or both, and which of them you want changes how the content is structured — prose for one, rows for the other.

• The video. With the table visible.
• What you want out. A document, a sheet, or both.
• Whether you can reach the source. Access matters more than quality.
• Whether the narration matters. It usually explains the table.
• How exact the figures must be. It decides the approach.

What do you type into Vizard Agent?

Ask Vizard Agent to hunt for the source before it starts transcribing. It is one extra sentence in the brief, and it can replace the entire reading job with a single download that returns exact values rather than recovered ones.

Prompt

Variants worth knowing:

• "Look for the address in the frame." The step that saves the work.
• "Column by column." How to read a table from pixels if you must.
• "Tell me which figures you are less sure of." Turns a risk into a list.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real video containing an on-screen spreadsheet. The turn comes at step five, where Vizard Agent stops reading pixels altogether and starts looking instead for the file those pixels came from.

1. Checks the video and samples its frames.
2. Pulls clearer frames to read the content.
3. Reads the main scenes at higher resolution.
4. Crops the table region to read it column by column.
5. Reads the sheet's address where it appears on screen.
6. Zooms in on the link to identify it exactly.
7. Checks whether the original file can be downloaded directly.
8. Fetches the data from the identified source.
9. Checks whether the narration should be included in the document.
10. Builds the document and the spreadsheet from the recovered content.

Step four is the right technique if you are stuck with the pixels. Reading a screenful at once produces mush; cropping to one column and reading it down gives the recognition a consistent shape to work with.

Step five is worth trying every time. People screen-record their own sheets, and the address is right there in the browser bar — so the exact data is often one fetch away from a video that looks like it needs transcribing.

Step nine is what makes the document useful rather than merely accurate. A table of figures without the explanation that ran over it is a spreadsheet nobody can interpret in six months.

What does the result look like?

Two files rather than a video. A written document carrying the narration with its timings, and a spreadsheet carrying the rows — and Vizard Agent tells you whether the numbers came from the real file or were read from the screen.

When does this not work well?

Recovery from pixels has a hard floor. Small figures, heavy compression, a table that scrolls quickly, or a screen photographed rather than recorded will each produce values that look perfectly plausible and are not, and Vizard Agent has to flag them rather than fix them.

• Small text. Below a certain size nothing is reliable.
• Heavy compression. Digits blur into each other.
• Fast scrolling. Rows pass before they resolve.
• A screen filmed on a phone. Moire and angle make it worse.
• No visible source. Then the pixels are all there is.

How do you fix a result that came back wrong?

Point Vizard Agent at the cell. It knows which of the figures came out of the real file and which were read off the screen, so a correction of this kind usually just confirms a value it had already flagged to you as uncertain.

• "That figure is wrong." The frame is re-read at that timestamp.
• "The columns are misaligned." The table region is re-cropped.
• "I need the narration too." The transcript is added with its timings.
• "Can you get the real file?" The on-screen address is hunted for again.

How does Vizard Agent compare to doing it yourself?

By hand you pause the video and type the table out, which takes an hour and introduces exactly the errors you were worried about. Nobody thinks to look at the browser bar in the frame, where the link to the actual sheet has been sitting the whole time.

Common questions

Why look for a link first? Because the real file gives exact values. Vizard Agent tries that before reading pixels.

What if the sheet is private? Then Vizard Agent falls back to reading the screen and says which figures are uncertain.

Can it produce both formats? Yes — a document for the words and a spreadsheet for the rows.

How accurate is reading from screen? Good at large sizes, unreliable at small ones. Vizard Agent tells you which is which.

What about a table that scrolls? Vizard Agent reads it at successive timestamps and stitches the rows together.

Does it include the narration? If you ask. Vizard Agent adds it, because it usually explains what the table means.

Will the timings be kept? Yes, in the document, so you can find the moment again.

What if the video shows a chart rather than a table? Values read off a chart are estimates. Vizard Agent will say so.

Can it check the totals? Yes, and a column that does not add up is a good sign of a misread digit.

What about a screen filmed on a phone? Much harder. Vizard Agent will tell you what it could not resolve.

Can it find the file if only part of the link shows? Sometimes, by zooming in on the frame where the most of it is visible.

Does it need the whole video? For a scrolling table, yes. For a static one, the frames it appears in.

Can I get a summary too? Yes. Ask Vizard Agent for the summary alongside the raw rows.

Is this different from getting on-screen text? The technique overlaps; the difference is chasing the source file first.