You have a folder of 200 receipts and your finance team wants them in a spreadsheet by Friday. Each receipt has a vendor name, date, total amount, and tax. But they are all images. PDFs, photos, screenshots. Not text.
Usually this means one of three things. Type them out by hand. Pay for a separate OCR tool. Or build a Python script that calls a vision API. All three cost time, money, or both.
=GPTV() takes a different approach. It takes an image URL and a question. It sends the image to an AI vision model. And it puts the answer right in your cell as text. One formula per image. Drag it down a column and 200 receipts get processed the same way 200 =SUM() formulas would.
Your first image analysis
=GPTV("https://example.com/receipt.jpg", "Extract the vendor name, date, and total amount from this receipt. Return as: Vendor, Date, Total")Sends the image to an AI vision model and returns structured text. The second argument tells the model exactly what to extract.
The first part is the image URL. The second part is your prompt, telling the AI what to do with the image. If you skip the prompt, =GPTV() just describes what it sees.
=GPTV(A2)With no prompt, returns a general description of the image at the URL in A2. Useful for quick image inventorying.
=GPTV() works with any publicly accessible image URL, including JPG, PNG, WebP, and PDF (first page). It supports OpenAI (GPT-4o vision), Google Gemini, and Claude depending on your selected provider in Extensions > SheetMagic > Settings.
Receipt processing at scale
This is the use case that pays for itself the fastest. Every receipt image becomes a row of structured data. No typing needed.
The setup
| Column A | Column B | Column C | Column D | Column E |
|---|---|---|---|---|
| Receipt URL | Vendor | Date | Amount | Category |
https://...receipt1.jpg | (formula) | (formula) | (formula) | (formula) |
Option 1: One formula per field
=GPTV(A2, "What is the vendor or merchant name on this receipt? Return only the name, nothing else.")Extracts just the vendor name. Clean, single-value output that is easy to filter and sort.
=GPTV(A2, "What is the date on this receipt? Return in YYYY-MM-DD format. Return only the date.")Extracts the date in a consistent format. Specify the format you want and the AI follows it.
=GPTV(A2, "What is the total amount on this receipt, including tax? Return only the number with currency symbol.")Extracts the total. Asking for 'only the number' prevents the AI from adding explanatory text.
=GPTV(A2, "Categorize this receipt as one of: Office Supplies, Travel, Meals, Software, Equipment, Other. Return only the category.")AI categorization based on the receipt contents. Not a keyword match, but actual understanding of what was purchased.
Option 2: All fields in one call with AILISTH
If you want to use fewer tokens, you can pull everything out in a single call and spread it across columns:
=AILISTH(GPTV(A2, "Extract from this receipt: 1) Vendor name, 2) Date in YYYY-MM-DD format, 3) Total amount with currency, 4) Category (Office Supplies/Travel/Meals/Software/Equipment/Other). Return only the 4 values, no labels."))GPTV extracts the data, AILISTH splits it into four separate cells (B2 through E2). One API call instead of four.
Option 2 uses fewer AI tokens but is harder to troubleshoot when something goes wrong. Start with Option 1 (one formula per field) while you get your prompts right. Then switch to the combined version once you trust the output.
Product image categorization
E-commerce teams deal with thousands of product images that need tags, descriptions, and categories. =GPTV() turns this from a manual review job into a formula.
Auto-tagging product images
=GPTV(A2, "List 5 descriptive tags for this product image. Tags should be useful for e-commerce search: material, color, style, use case, category. Return as comma-separated values.")Generates search-friendly tags from the image itself. No product data needed. The AI sees the product.
Writing product descriptions from photos
=GPTV(A2, "Write a 2-sentence e-commerce product description based on this image. Focus on material, color, and key features visible in the photo. Professional tone.")Creates product copy directly from the image. Combine with existing product data for richer descriptions.
Quality control checks
=GPTV(A2, "Evaluate this product image for e-commerce use. Check: 1) Is the background clean/white? 2) Is the product centered? 3) Is the lighting even? 4) Any visible defects? Rate overall quality 1-10. Return a brief assessment.")Automated QA for product photography. Flag images that need reshooting before they go live.
Put your image URLs in column A
Upload your product images to Google Drive, your CDN, or any hosting service. Put the public URL for each image in column A, one per row.
Write your tagging formula in B2
Use the auto-tagging formula above. Test it on one row first. If the tags are too generic ("product, item, thing"), make your prompt more specific. Mention the product category so the AI has context.
=GPTV(A2, "List 5 descriptive tags for this clothing product image. Include: garment type, color, material, style (casual/formal/athletic), and season. Comma-separated, no numbering.")Adding category context (clothing) produces more relevant tags than a generic prompt.
Add description and QA formulas
Put the product description formula in C2 and the quality check formula in D2. Each column runs on its own. You can drag them all down at once.
Drag down and review
Select B2:D2, grab the fill handle, and drag down to cover all your products. Check the first 10 rows to make sure things look right. Then let the rest process.
Screenshot analysis and documentation
Screenshots are everywhere, but nobody treats them like data. Bug reports, competitor research, design reviews. They all live as images that someone has to look at by hand. =GPTV() makes them searchable.
Bug report triage
Your QA team sends screenshots of UI problems. Instead of reading through each one, pull out the key details automatically:
=GPTV(A2, "Analyze this screenshot of a software bug. Describe: 1) What screen/page is shown, 2) What appears to be wrong, 3) Severity estimate (Critical/High/Medium/Low). Keep each answer to one sentence.")Turns a screenshot into structured bug metadata. Useful for triaging a backlog of visual bug reports.
Competitor research
Take screenshots of competitor websites, pricing pages, or product features and analyze them at scale:
=GPTV(A2, "This is a screenshot of a competitor SaaS pricing page. Extract: 1) Number of pricing tiers, 2) Price range (lowest to highest monthly), 3) Key differentiating feature of each tier. Be concise.")Competitive intelligence from screenshots. No need to manually review each competitor page.
Design review feedback
=GPTV(A2, "Review this UI design screenshot. Identify: 1) The primary user action on this screen, 2) Any potential usability issues, 3) Accessibility concerns (contrast, text size, touch targets). Be specific.")Automated design review checklist. Not a replacement for a designer, but a useful first pass.
Document extraction
PDFs, scanned documents, and photographed paperwork all work with =GPTV(). The AI vision model reads the document the same way it reads any image.
Invoice data extraction
=GPTV(A2, "Extract from this invoice: 1) Invoice number, 2) Invoice date, 3) Due date, 4) Vendor name, 5) Total amount due. Return only the values, comma-separated.")Processes invoice images or PDF URLs into structured data. Handles varying invoice formats automatically.
Business card scanning
=GPTV(A2, "Read this business card image. Extract: 1) Full name, 2) Job title, 3) Company, 4) Email, 5) Phone number. If any field is not visible, return N/A.")Digitizes business cards from photos. The N/A fallback keeps your data consistent.
Handwritten notes
=GPTV(A2, "Transcribe the handwritten text in this image. Maintain the original structure (bullet points, numbered lists) where visible. If any text is illegible, mark it as [illegible].")OCR for handwritten notes. Works best with reasonably clear handwriting. Not a replacement for typed text, but a solid first pass.
=GPTV() processes the first page of multi-page PDFs. If your document has more than one page, split it into individual page images first. Or use one row per page in your spreadsheet.
Tips for accurate image analysis
Be specific about the output format. "Extract the date" gets you "The date on the receipt appears to be March 15, 2026." But adding "Return only the date in YYYY-MM-DD format" gets you "2026-03-15." The second one is what you want in a spreadsheet.
Give context in your prompt. "Describe this image" gives generic output. "This is a product photo for an outdoor furniture catalog. Describe the material, color, and style" gives useful output. More context means better results.
Use fixed categories. When you want classification, give the AI a set list of options: "Categorize as one of: A, B, C, or D." Open-ended categories lead to inconsistent labels across rows.
Handle missing data on purpose. Add "If [field] is not visible, return N/A" to your prompt. Without this, the AI might guess or leave cells blank. That breaks your data.
Test on 3 to 5 images before scaling. Image quality, angles, and formats vary. What works on your first test receipt might struggle with a blurry photo. Get your prompt right on a small sample, then drag down.
Getting started
You can try all of this on the free tier. No credit card needed. You get 3,000 AI tokens, enough to analyze a handful of images and see how =GPTV() fits your workflow. No API keys to set up. No configuration.
Try it now. Install SheetMagic free, paste any image URL into cell A1, and type this in B1:
=GPTV(A1, "Describe what you see in this image in 2 sentences")
Your first AI image analysis will show up in seconds.
When you are ready to process images at scale, paid plans start at $19/month with 3 million AI tokens. See the pricing page for details.
For the complete formula reference, read the complete guide to AI in Google Sheets. To see how =GPTV() fits alongside text generation and list extraction, check the full AI formulas guide. And for a comparison of building this with custom code vs. an add-on, see Custom Functions vs Add-ons vs API (coming soon).
Have questions? visit the FAQ for answers about AI formulas, image analysis, and billing.