Getting a Table Out of a Screenshot

A table in a screenshot can be turned into an Excel sheet by an AI model that reads the image and returns the rows and columns. A clean screenshot of a table on screen converts well; an angled phone photo, handwriting or a small, blurry image often does not. This article covers what works, what does not, and one fact to know before uploading: the image is sent to Google's Gemini API to be read.

How does AnyToExcel read a table from a screenshot?

AnyToExcel reads a table from a screenshot by sending the image to a Google Gemini model with a fixed instruction: return one object per row, use the first row as column names, write an empty cell as an empty string, use ? for any value it cannot read, and never guess a number. The rows that come back go through the same steps as a JSON file, so each column is then tested for numbers and dates before the .xlsx is written.

• Accepted images: PNG, JPEG and WebP (.png, .jpg, .jpeg, .webp), up to 4 MB.
• Two ways in: upload the image on the home page converter, or press Ctrl+V on the home page to paste a screenshot straight from the clipboard.
• Taking the screenshot: Win+Shift+S on Windows, or Cmd+Ctrl+Shift+4 on a Mac, copies a selected area to the clipboard.
• No account is needed, and an image counts as one conversion like any other file.

Which screenshots convert to Excel well?

Screenshots convert to Excel well when the table fills most of the image, the text is sharp, and the rows and columns are lined up straight. A screenshot of a table on a web page, in a PDF viewer, in an email or in another app's report screen is the best case, because the text is printed by a computer and nothing is tilted.

If the table is too wide to read in one screenshot, zoom the page so the table fills the screen width before taking the screenshot. A larger screenshot of part of the table converts better than a tiny screenshot of all of it.

What happens to numbers the model cannot read?

Numbers the model cannot read are meant to come back as ? rather than as a guessed value, because the model is told to write ? for any unreadable value and never to guess a number. A ? in a column of figures also keeps that whole column as text in the .xlsx, since ? is not a number, which makes the gap visible instead of turning it into a wrong total.

That instruction is not a guarantee. An AI model can still misread a value it believes it has read correctly, for example a 6 as an 8 in a blurred image. Check the totals and a sample of rows against the original before using the sheet for anything that matters, such as accounts or invoices.

In that example the Q2 column is written as numbers. The Q1 column is written as text because of the ?, so a SUM over Q1 returns 0 until the missing figure is typed in from the source.

Where is a screenshot sent when it is converted?

A screenshot converted by AnyToExcel is sent to Google's Gemini API, which reads the image and returns the table. Images are the only exception: CSV, JSON, HTML and PDF files are processed on AnyToExcel's own server and are not sent to any third party.

AnyToExcel does not store the image or the converted file. The image is passed to Google in the conversion request and the .xlsx is returned in the same response. How Google handles the image is set by Google's terms for the Gemini API, not by AnyToExcel. Do not upload a screenshot of a table you are not allowed to share with Google, such as one containing customer or patient data.

Is there a better option than a screenshot of a table?

There is a better option than a screenshot of a table whenever you can get the table as text, because text is converted exactly, character for character, while an image has to be read by a model. All of these are processed on AnyToExcel's own server:

• A table on a web page: select it, copy it and paste it into the paste box on the home page, or save the page and convert it with the HTML table to Excel converter.
• A table in a PDF: upload the PDF itself. A PDF with selectable text is converted from its text, one sheet per page that holds a table. A scanned PDF has no text and is rejected, so for a scan a screenshot is the only route.
• A table in an email or a document: copy the rows and paste them into the paste box on the home page.

Why did my screenshot fail to convert?

A screenshot fails to convert when the model finds no table in it, when the columns it returns cannot be lined up into rows, when the image is over 4 MB, or when the image service is busy. Each case returns its own message:

Image formats other than PNG, JPEG and WebP, such as HEIC, GIF and TIFF, are not read as images. Convert them to PNG or JPEG before uploading.