=GPT_VISION() — ask questions about an image
=GPT_VISION() sends an image URL to a vision model and returns the answer to your question about it — read text from a screenshot, describe a product photo, extract fields from a scanned document. Part of GPT for Sheets.
Syntax
=GPT_VISION(url, prompt)
| Parameter | Required | Description |
|---|---|---|
url |
yes | A link to the image, e.g. "https://example.com/someimage.png" — or a cell with the link. |
prompt |
yes | What to do with the image, e.g. "extract First name and Last name". |
Examples
Write alt text for a column of product image URLs in A2 (e-commerce SEO):
=GPT_VISION(A2, "Write a 10-word alt text describing this product photo")
Extract names from scanned business cards:
=GPT_VISION(A2, "extract First name and Last name")
Pull the total from receipt photos for expense tracking:
=GPT_VISION(A2, "Return only the total amount and currency")
Audit competitor screenshots — what’s the main call to action?
=GPT_VISION(A2, "What is the primary call-to-action button text on this page?")
Common errors
- The image URL must be publicly reachable — links behind a login (including private Google Drive files) fail.
- Point the URL at the image file itself, not at an HTML page that displays it.
- Low-resolution photos read poorly; if extraction misses text, try a larger version of the image.
Related functions
- =GPT_CREATE_IMAGE() — generate images instead of reading them
- =GPT_EXTRACT() — extract values from text
- =GPT() — text-only prompts
Try it
=GPT_VISION() ships with GPT for Sheets — no API keys needed. Install the add-on from the Google Workspace Marketplace.