Google Cloud Vision API reads text out of an image extremely well, and then it stops. It returns the words and where they sat on the page, with no merchant, no date, no tax, no total, and no line items, because Vision has no receipt or invoice parser at all. Google's own OCR documentation tells you to leave for Document AI if you want structured fields. ReceiptOCR gives you those fields directly, in a browser, and exports Excel or CSV. Upload a receipt below and see the difference between text and data.
Upload your receipts and invoices
Drop files here or click to upload
Up to 50 files
Uploading...
Cloud Vision is a superb OCR engine and a poor receipt tool, and those two statements do not conflict. It was built to find text in images, not to understand a receipt. Everything that makes a receipt useful to accounting is left for you to construct.
Vision's feature list covers text, faces, labels, logos, landmarks, and objects. There is no receipt parser, no invoice parser, and no field extraction of any kind. The response has no total key because Vision never looks for one.
The JSON gives you pages, blocks, paragraphs, words, and symbols with bounding boxes. That hierarchy is spatial, not meaningful. Turning "$42.18" into the total, rather than one of nine dollar figures on the page, is your problem to solve.
There is no CSV or Excel output. For a PDF or TIFF, the result is not even returned inline: Vision writes a JSON file into a Cloud Storage bucket you have to create and own first.
The first 1,000 units a month are free, but Google requires a Cloud project with billing enabled before the API will answer at all. The free tier is a discount on a billed account, not a way in without a card.
ReceiptOCR does the part Vision leaves undone. It reads the document, decides which number is the total, which date is the transaction date, and which lines are line items, then hands you a spreadsheet.
Merchant, date, subtotal, tax, tip, and total arrive labeled. No regex over raw text, no guessing which of the nine dollar amounts on a receipt is the one you want.
Download the extracted documents as a spreadsheet, a CSV, or a QuickBooks-ready file. Nothing gets written to a bucket you then have to go and fetch.
No GCP project, no gcloud CLI, no service-account credentials, no enabled API. Open a browser, upload the documents, take the file.
Each line with description, quantity, and unit price where the document carries them. Vision can see those words; it cannot tell you they form a line item.
Every field is on screen and editable before the data leaves, so a misread total gets caught by a person rather than landing silently in your books.
The same fields come back as JSON over a REST endpoint, already parsed into meaning, so you skip the layout heuristics entirely.
From a document to a spreadsheet without writing layout heuristics.
Receipts, invoices, or both in the same batch. Photos, scans, and PDFs go in together, and nothing has to be written to a storage bucket first.
Tip: Try a receipt with several dollar amounts on it. Telling the total apart from the subtotal, the tax, and the change due is exactly the job raw OCR leaves to you.
Review merchant, date, tax, total, and every line item on screen. Correct anything the engine flagged before it becomes a row in your books.
Download Excel, CSV, or a QuickBooks-ready file, or call the REST API and receive the same fields as JSON inside your own workflow.
Built for US teams that need receipt and invoice fields in a spreadsheet, not a transcript of every word on the page.
You need the total, not every word on the receipt. Raw OCR text is more work than typing it in, which defeats the entire point of automating.
You wired up Vision, got beautiful text, and discovered the hard part was never the OCR. Writing regex and layout rules per merchant is where the project stalled.
Supplier invoices in dozens of layouts headed for a ledger. Text with coordinates is not a payable; vendor, amount, and due date are.
A few hundred documents a month and no engineer to build the semantic layer that Vision deliberately does not include.
Last updated July 2026.
Google Cloud Vision API is one of the best OCR engines available, and it is the wrong tool for reading a receipt. Those two things are both true, and understanding why saves you a wasted sprint. Vision converts an image into text and tells you where each word sat. It does not tell you which number is the total, and it never will, because it has no receipt parser. Google's own documentation says as much.
Text and coordinates. TEXT_DETECTION returns the full extracted string plus individual words with bounding boxes. DOCUMENT_TEXT_DETECTION, the dense-document mode, returns a fuller structure organized as pages, blocks, paragraphs, words, and symbols, each with its own bounding box. That hierarchy is geometric and typographic, not semantic. There is no merchant key, no date key, no tax key, no total key, and no line items array anywhere in the response, because Vision is not looking for those concepts. It is telling you what characters appeared on the page and where.
No. Vision's published feature list covers text detection, document text detection, landmark, logo, label, image properties, object localization, crop hints, web entities, SafeSearch, and face detection. There is no receipt parser, no invoice parser, no expense parser, and no field extraction of any kind. This is the single most important fact on this page. If you are evaluating Vision to pull totals off receipts, the feature you are looking for does not exist in the product.
Google answers this in its own OCR documentation, which carries a notice telling readers that if they are detecting text in scanned documents, they should try Document AI for optical character recognition, structured form parsing, and entity extraction. That sentence is the whole distinction. Vision does OCR. Document AI does OCR plus structured parsing plus entity extraction, and its Expense and Invoice parsers are the products that actually return merchant, date, total, and line items.
The pricing reflects it exactly. Raw OCR costs $1.50 per 1,000 pages whether you buy it from Vision or from Document AI's Enterprise Document OCR. The Document AI Invoice and Expense parsers cost $10 per 1,000 pages. That roughly seven-fold gap is the price of meaning: you are not paying more for better character recognition, you are paying for the model that knows a receipt has a total.
These figures come from Google's own Vision pricing page in July 2026. Both text detection features are priced identically, and a unit is one feature applied to one image. For a PDF or TIFF, each page counts as one unit.
| Units per month | Price | What you get |
|---|---|---|
| First 1,000 | Free | Raw text and bounding boxes |
| 1,001 to 5,000,000 | $1.50 per 1,000 | Raw text and bounding boxes |
| Above 5,000,000 | $0.60 per 1,000 | Raw text and bounding boxes |
| Document AI Expense or Invoice parser | $10 per 1,000 pages | Named fields and line items |
The cheap tier is cheap because it does less. Budgeting Vision at $1.50 per 1,000 pages and then discovering you need a developer to write and maintain the parsing layer is how a low unit price turns into an expensive quarter. Price the engineering, not just the API call.
No. Vision returns JSON, and there is no CSV or Excel output anywhere in the product. For PDF and TIFF input the result is not even returned to you inline: Vision writes a JSON file into a Cloud Storage bucket that you must create and own. So the path from a scanned invoice to a spreadsheet runs image, JSON in a bucket, code you wrote to fetch it, code you wrote to interpret it, and finally a row. That is four steps that are your responsibility.
The first 1,000 units per feature per month are free, and that allowance recurs monthly rather than expiring like a trial credit. But you still need a Google Cloud project with billing enabled before the API will respond, so a payment method is required even to use the free units. Add the gcloud CLI, credentials, and a client library, and the free tier is free in dollars and expensive in hours.
Choose Vision when text is genuinely what you want. Digitizing books, reading signs or license plates from photographs, transcribing handwriting, indexing scanned archives for full-text search, or extracting text in dozens of languages: Vision is excellent at all of it, cheap at scale, and backed by Google's research. If your job ends when the characters are recognized, it is a first-rate choice and we are not going to pretend otherwise.
Choose it, too, as a component when you already own the semantic layer. If you have built an extraction pipeline and only need a strong OCR engine behind it, Vision is a sensible, well-priced piece of infrastructure.
If the job is receipt and invoice data, the OCR is only the first inch of it. For expenses, receipt OCR software reads receipts into labeled fields in a browser and the receipt to Excel converter lands them in a spreadsheet. For vendor bills, invoice OCR software handles header fields and line items. Large batches run through the bulk receipt scanner. Developers who want the parsed fields as JSON, without writing the layout heuristics, can call the receipt OCR API or the invoice OCR API. If you are comparing the cloud vendors, the Google Document AI alternative, AWS Textract alternative, and Azure Document Intelligence alternative pages cover per-page billing on each, and the OCR software pillar explains the difference between raw OCR, template OCR, and AI extraction in one place.
No. Vision returns raw text with bounding boxes, organized as pages, blocks, paragraphs, words, and symbols. There is no merchant, date, tax, total, or line items field in the response, and Vision has no receipt or invoice parser at all. Identifying which number is the total is code you write on top of the text.
Vision does OCR. Document AI does OCR plus structured form parsing and entity extraction, and Google says exactly that in its own Vision documentation. Document AI has Invoice and Expense parsers that return named fields and line items; Vision has none. Vision OCR runs $1.50 per 1,000 pages, while the Document AI parsers cost $10 per 1,000.
As of July 2026, the first 1,000 units a month are free, then $1.50 per 1,000 units up to 5 million, and $0.60 per 1,000 above that. A unit is one feature applied to one image, and each page of a PDF or TIFF counts as one unit. Text detection and document text detection are priced the same.
No. Vision returns JSON only. For PDF and TIFF input it does not even return results inline, writing a JSON file into a Cloud Storage bucket you must create and own. Getting from that to a spreadsheet is entirely code you write and maintain.
The first 1,000 units per feature per month are free and recur monthly, but you still need a Google Cloud project with billing enabled before the API responds. A payment method is required even for the free units, and you also need the gcloud CLI, credentials, and a client library.
A tool that returns named fields rather than raw text. ReceiptOCR reads receipts and invoices, labels merchant, date, tax, total, and line items, shows them for review, and exports Excel, CSV, or a QuickBooks-ready file, with a REST API when you want one and no Google Cloud project to set up.
You can, and many teams start there. Vision handles the character recognition well, but you then write the layer that decides which of several dollar amounts is the total, which date is the transaction date, and which lines are line items, and you maintain it as merchant layouts change. That semantic layer, not the OCR, is where these projects usually stall.
Google's Invoice and Expense parsers, compared.
Microsoft's prebuilt receipt and invoice models, compared.
Amazon's cloud OCR API, compared honestly.
The three types of OCR, and which one you need.
Parsed fields as JSON, without the layout heuristics.
Turn a pile of receipts into a clean spreadsheet.
