Your First AI Extraction:
What Happens When You Upload a Photo
Someone told you AI can pull data out of documents. You want to try it. But you're not sure if you need to learn Python first, or watch a video, or sign up for something that asks for your credit card. You don't. What you need: one photo of an invoice on your phone, and a browser tab. That's it. Below is exactly what happens when you do it — and you can follow along right on this page.
Key Takeaways
- You don't need Python, a credit card, or a sign-up to extract data from an invoice — the entire setup is one photo and three column names typed into a browser.
- The AI reads your document the way you do — it knows the number next to "Total Due" is what you owe, even if that number moves to a different corner on the next invoice.
- The entire journey from phone photo to finished spreadsheet takes 10 seconds — less time than you just spent deciding whether this was too complicated for you.
Before You Start: What You Need
Three things. If you have a phone and a browser, you already have everything.
A photo of a document. An invoice works best for this first try — the kind you'd get as a PDF attachment or snap with your phone. Any invoice. It doesn't need to be a perfect scan. A photo taken with your phone in normal lighting is fine.
A browser. Chrome, Safari, Firefox — doesn't matter. You don't need to install anything. No app store. No Chrome extension. No download.
The names of three things you want to find. For an invoice, pick: Date, Vendor, and Total. That's it. You don't need to know what a "field" is. You don't need to draw boxes. Just the three words.
You don't need to create an account to try this. The demo below works anonymously — no email, no password, no credit card. If it works for you, you can sign up later. If it doesn't, you lost 60 seconds.
Step by Step: Your First Extraction
Six steps. You'll be done before you finish reading this section.
Here's the whole thing. No hidden steps. No "click through five configuration screens first." Just this:
That's the entire workflow. If you've read this far, you've spent more time reading than it takes to do it. Try it now:
Files are processed securely and not stored.
What Just Happened?
The AI didn't scan your document for coordinates. It read it — the way a person would.
Traditional OCR works by finding dark shapes on a page and matching them to letters. It doesn't know the difference between "Invoice #1042" and "1042 Main Street." Both are characters. When a vendor changes their layout — moves the invoice number to a different corner — position-based extraction breaks because it was looking at coordinates, not meaning.
Custom Column Extraction works differently. You type what you want — "Date," "Vendor," "Total" — and the AI reads the entire document to find values matching those meanings. It looks at the invoice number label and knows the value next to it belongs in that column, regardless of where they sit on the page. The column names you typed become the headers of your spreadsheet. That's the whole mechanism: you define the output, the AI understands the input.
This is why you didn't need to configure anything. No templates to build per vendor. No training samples to upload. No boxes to draw around fields. The AI reads documents the way you do — by understanding what they say, not where the pixels are. For a deeper look at why this eliminates setup entirely, see our article on document extraction without setup.
The gap between "text recognized" and "data usable" used to be filled with your time — copying each value from an OCR text dump into the right spreadsheet cell. AI extraction closes that gap. You get structured columns, not raw text. If you're new to the concept, what is AI data entry explains the difference in detail.
You just did in 10 seconds what manual typing takes 3 minutes per page. That's an 18x difference. The reason isn't faster typing — it's that the AI skips the "find it, verify it, type it" loop entirely.
Frequently Asked Questions
Does this work with handwritten documents?
Yes. The AI reads handwriting the same way it reads printed text — by understanding the visual patterns. A handwritten invoice or receipt works. Messy handwriting may reduce accuracy, but the system handles cursive and hand-printed text without special configuration. You don't enable a "handwriting mode" — it just handles it.
Do I need to create an account?
No. The demo above works anonymously. You can upload and process files right now, on this page, without entering an email or password. An account gives you higher limits, batch processing, and export options — but for your first try, you don't need any of that. Just upload and go.
Does it work with PDFs?
Yes. PDF, JPG, PNG, WebP, and screenshots all work. Multi-page PDFs work too — you can select which pages to extract from. If you have a scanned document saved as a PDF, just drag it in like you would a photo.
Can I extract more than three columns?
Yes. Three columns — Date, Vendor, Total — is the minimal starting point. You can type as many column names as you need: Invoice Number, Due Date, Subtotal, Tax, Line Items, whatever appears on your document. You can also define Computed Columns that calculate values during extraction — for example, Line Total (Qty × Unit Price) — and Inferred Columns where the AI categorizes data using options you provide, like Category (options: Meals/Transport/Office/Other).
How accurate is it?
Up to 99% for printed document data. Accuracy depends on image quality — a well-lit photo of a clean document will produce near-perfect results. A dark, blurry, angled photo of a crumpled receipt will be less reliable. The same factors that would make it hard for a person to read. The difference is that the tool processes a page in 5–10 seconds, so even if you need to verify a couple of fields, you're still saving 95% of the time you'd spend typing the whole thing.
What happens to my files?
Files uploaded through the demo are processed and then automatically deleted. They are not stored, not used for training, not accessible to anyone but you during the processing session. For account users, files are retained in your history until you delete them, and you can clear your history at any time.
What if I have more than one document?
The tool is built for batch processing. Upload multiple files at once — 5 invoices, 20 receipts, 50 purchase orders — and they process together, merging into a single spreadsheet with consistent column names. No per-document setup. Every file goes through the same extraction logic you defined with your column names, regardless of how different the layouts are.
You just did your first AI extraction. You took a concept that sounded like it required technical knowledge, and you did it in six steps. The mental barrier — "this is probably too complicated for me" — was the only real obstacle. Now you know what actually happens. Try it again with your own file, or with a batch of files. Same steps. Same result. No new learning curve.