Receipt OCR (optical character recognition) is software that reads the text on a photographed receipt and turns it into structured expense data — merchant name, amount, date — so you don't have to type it yourself. It's the difference between filing a receipt in five seconds and filing it in sixty.

How accurate is receipt OCR?

On a crisp, printed receipt photographed in decent light, modern OCR reads the merchant and total reliably. Accuracy drops on faded thermal paper, handwriting, and unusual layouts. That's why Starlog treats OCR output as a draft: every field stays editable, and the original image is always kept as the source of truth.

Does Starlog's OCR run on my phone or in the cloud?

On your phone. Starlog uses Google ML Kit's on-device text recognition, so the receipt image isn't sent to a third-party OCR service for reading. The image is then backed up to your own Google Drive — your storage, not a vendor database.

Does Starlog read individual line items?

No — and that's deliberate. Reading that a receipt totals ₹2,400 is reliable; itemising fifteen lines with their taxes is where OCR gets subtly wrong. Starlog drafts the store and total, you confirm, and the full image is preserved for any detail a person ever needs to check.

Does it work on PDF receipts?

Yes. For PDF receipts — email invoices, ride-hailing receipts, online orders — Starlog extracts the first page and runs the same on-device OCR over it.

What happens when the OCR misreads something?

You see it and fix it in a tap — the pre-filled fields sit in an editable form before anything lands on your books. Nothing is auto-committed, and the original image stays in your Drive, so even a mistake you miss today can be settled against the source later.

Receipt OCR

Receipt OCR, without the magic act.

What OCR actually reads off a receipt, where it still fails in 2026, and how Starlog is built so a misread never sneaks into your books.

Last updated: 2026-06-11

Every receipt app promises some version of “just snap a photo and we extract everything automatically.” It's a good pitch because it's mostly true — and the “mostly” is exactly where your bookkeeping either holds up or quietly goes wrong.

This page explains what receipt OCR is, what Starlog's OCR does and deliberately doesn't do, and how to get the most out of it. No claimed accuracy percentages, no “AI magic” — just how the thing actually works.

What is receipt OCR?

Receipt OCR (optical character recognition) is software that reads the text on a photographed receipt and turns it into structured expense data — merchant, amount, date — so you don't type it yourself. It's what makes capturing an expense a five-second job instead of a sixty-second one.

Where it runs matters more than most people realise. Many apps upload your receipt to a cloud OCR service to read it — which means a copy of every supplier, price, and purchase passes through someone else's infrastructure. The alternative is on-device OCR: the photo is read by your phone itself, and never leaves it for the sake of text extraction.

What Starlog's OCR reads — and what stays yours

Starlog uses Google ML Kit's text recognition, running entirely on your phone. When you snap a receipt:

01The store name and total amount are read off the image and pre-filled, usually in a second or two.
02The date, category, and notes are yours to set — a tap each, from your own category list.
03Everything lands in an editable form before it touches your books. The autofill is a starting point, not a verdict.

PDF receipts work too: Starlog extracts the first page and runs the same on-device recognition over it — handy for email invoices and ride-hailing receipts.

On your phone, not on our servers

Because recognition happens on-device, Starlog doesn't ship your receipt to a third-party OCR API to read it. And once captured, the original image is filed into your own Google Drive — sorted by business, year, and month — not into a vendor's database you'd have to export your way out of someday.

That's a deliberate position: the extracted text is a convenience, but the image is the record — the thing an audit asks for in three years. Records like that belong in storage you control. More on the full posture on the security & privacy page.

Where receipt OCR still fails

An honest list, because real receipts aren't the crisp ones in the demos:

Faded thermal paper — a slip that sat in a hot car for a week can be half-gone, and OCR can't read what isn't there.
Handwriting — scrawled tips and market-stall totals read far less reliably than print.
Odd layouts — every point-of-sale system formats differently, and “amount due” vs “amount paid” vs “change” can confuse extraction.
Line items — reading the total is reliable; itemising fifteen lines and their taxes is much easier to get subtly wrong.
Format ambiguity — is 03/04 March or April? Is 1.500 one-and-a-half or fifteen hundred? Confident extraction can quietly pick wrong.

The full breakdown is in What OCR still gets wrong on real receipts.

Fail visibly, never silently

There are two ways for OCR to be wrong, and they're not equally bad. Failing visibly — “can you check this total?” — costs you a glance. Failing silently — confidently reading ₹240 from a receipt that says ₹2,400 — sails into your records looking finished, and surfaces months later as a number that won't reconcile.

Starlog is designed around that asymmetry: OCR drafts, you confirm, corrections take a tap, and the original image is never thrown away. A human stays in the loop by design — which is the correct division of labour for records you're legally responsible for.

Getting better reads

OCR quality is mostly photo quality. Three habits cover most of it:

01Capture early — photograph thermal receipts the day you get them, before heat and light eat the ink.
02Fill the frame — flat receipt, square-on angle, even light, no thumb over the total.
03Long receipts: prioritise the header and the total in one clean shot rather than cramming every line in.

The complete checklist is in How to photograph a receipt so the OCR actually works.

FAQ

What is receipt OCR?: Receipt OCR (optical character recognition) is software that reads the text on a photographed receipt and turns it into structured expense data — merchant name, amount, date — so you don't have to type it yourself. It's the difference between filing a receipt in five seconds and filing it in sixty.
How accurate is receipt OCR?: On a crisp, printed receipt photographed in decent light, modern OCR reads the merchant and total reliably. Accuracy drops on faded thermal paper, handwriting, and unusual layouts. That's why Starlog treats OCR output as a draft: every field stays editable, and the original image is always kept as the source of truth.
Does Starlog's OCR run on my phone or in the cloud?: On your phone. Starlog uses Google ML Kit's on-device text recognition, so the receipt image isn't sent to a third-party OCR service for reading. The image is then backed up to your own Google Drive — your storage, not a vendor database.
Does Starlog read individual line items?: No — and that's deliberate. Reading that a receipt totals ₹2,400 is reliable; itemising fifteen lines with their taxes is where OCR gets subtly wrong. Starlog drafts the store and total, you confirm, and the full image is preserved for any detail a person ever needs to check.
Does it work on PDF receipts?: Yes. For PDF receipts — email invoices, ride-hailing receipts, online orders — Starlog extracts the first page and runs the same on-device OCR over it.
What happens when the OCR misreads something?: You see it and fix it in a tap — the pre-filled fields sit in an editable form before anything lands on your books. Nothing is auto-committed, and the original image stays in your Drive, so even a mistake you miss today can be settled against the source later.