OCR PDF — make a scanned PDF searchable
A scanned PDF is a photograph of pages: it looks like a document, but no tool can search it, copy from it or convert its words — because as far as the file is concerned, there are no words in it, only pictures of them. OCR PDF reads those pictures and puts the words back as a real text layer, in your browser, free, with no sign-up.
Updated 21 November 2026 · Matches PdfLuxe v365
By the PdfLuxe team · Contact
The short answer
How do I OCR a PDF for free? Open OCR PDF, add your scanned PDF, pick the language printed on the pages and click Run OCR — the words are recognised on your own device and appear in the box on the right, each page with a confidence score. Download them as a plain .txt file, or as a searchable PDF: the same pages with the recognised text hidden inside, ready to search and copy. Everything below covers the 10-page run limit, the seven languages, the optional cloud engines, and how to get a cleaner read from a rough scan.
Steps
- Open OCR PDF and add your scanned PDF. If what you have is a photo of a page rather than a PDF, use “Or read a photo directly” instead: pick the JPG or PNG and its text is read in exactly the same way.
- Under “Language on the page”, pick the language printed on the pages — English is the starting choice, and Lithuanian, Hindi, Russian, German, Spanish and French are also there. Leave the OCR engine on “On your device — Tesseract (free + unlimited, nothing uploaded)”, and tick “Straighten a tilted page first” if the scan leans.
- Click Run OCR. One run reads up to 10 pages: each page is read three times with different settings and the clearest read is kept, and the recognised words appear in the box on the right, page by page, each page headed with its confidence percentage. Then download the words with Download .txt, or click Download searchable PDF for the same pages with the recognised text added as a hidden layer.
The honest limits, stated plainly. One run reads the first 10 pages of a PDF; if the document is longer, the rest are skipped and the result message says so — split a long scan into parts with Split PDF and OCR each part. The first run downloads the OCR engine and the language pack (a few MB), so it needs an internet connection once; after that, the reading itself happens on your device and nothing is uploaded. OCR reads print, not handwriting, and only as well as the scan allows: a straight, well-lit, sharp page reads far better than a blurred, skewed or shadowed one, and the confidence percentage beside each page tells you how much to trust it — the result flags any page that scored below 70%, so proofread anything that matters. The tool also offers cloud OCR engines (OCR.space, Google Cloud Vision and Azure Vision Read) that run on your own free key from AI Settings; those do send each page image to that service, and the searchable PDF they build carries the text as one hidden block per page, so search works per page rather than per word. Once a scan is searchable, the other converters can read it too — PDF to Word, PDF to Excel and PDF to Text all work on the words OCR added.
How the reading works
When you click Run OCR, each page is first drawn at high resolution — three times its normal size — because the recogniser reads small print far better from a big, sharp picture than from a small one. Two working copies are made of that picture: a grayscale copy with its contrast stretched, and a black-and-white copy produced by thresholding. Your original page is never altered; these copies exist only for the reading.
The page is then read three times, not once: twice on the grayscale copy with two different layout settings, and once on the black-and-white copy. Each read comes back with a confidence score, and the read with the highest score is the one that is kept — the grayscale reads usually win on clean scans, while the black-and-white read still wins on rough scans with grey or dirty backgrounds. If you ticked the deskew option, the engine first checks the page orientation and rotates the read copy upright before any of this.
The winning read provides more than the words: it provides where each word sits on the page. That is what makes the searchable PDF possible — your original page image stays exactly as it was, and each recognised word is written invisibly at its own position on top of it. Search lands on the right word, copying lifts the real text, and a page with nothing readable on it honestly scores 0% rather than a flattering guess.
Getting a cleaner read
OCR quality is decided mostly before the tool ever sees the file — by the scan itself. Four things make the biggest difference.
- Scan flat, straight and sharp. A page photographed at an angle, curled or blurred loses letters no engine can invent back. If a page leans, tick “Straighten a tilted page first” and let the engine rotate the read copy upright.
- Light it evenly. Shadows across the page and grey, dirty backgrounds are what the black-and-white read exists for, but an evenly lit page reads better from the start.
- Pick the right language. The language setting tells the engine which alphabet and word shapes to expect. A Lithuanian page read as English — or the other way round — fills with wrong letters, so match the setting to what is printed on the page.
- Mind the confidence scores. After a run, every page is listed with its percentage, and anything below 70% is flagged in the result. Re-scan those pages more carefully and run them again, rather than trusting a weak read.
On your device, or in the cloud
The default engine, Tesseract, runs entirely in your browser. The first run downloads the engine and the language pack for the language you picked — a few MB — and after that the pack is cached on your device, the reading happens locally, and no page image ever leaves your machine. It is free and unlimited, and it is the right choice for private documents: contracts, statements, medical letters, anything you would not post to a stranger's server.
Three cloud engines are offered as an alternative, each running on your own free key, which you paste into AI Settings: OCR.space (25,000 requests a month on its free key, no card), Google Cloud Vision (1,000 units a month free, card required) and Azure Vision Read (5,000 a month free on its F0 tier, card required). Those engines read each page on the provider's servers — the page image is sent to that service and the text comes back — and the searchable PDF they build carries the text as one hidden block per page, so search finds the page rather than the exact word. They can be worth trying on a scan the on-device engine reads weakly; the on-device engine remains the private default.
Privacy
With the default on-device engine, your PDF never leaves your device: it is opened in your browser, read in your browser, and the .txt and the searchable PDF are built in your browser and handed straight back as downloads. Nothing is uploaded, nothing is stored on a server, and there is no account, because there is nothing on a server to sign in to. The only downloads involved are the OCR engine and the language pack coming to you on the first run.
The cloud engines are the one exception, and the tool says so where you choose them: pick one, and each page image is sent to that provider under your own key, under that provider's terms. If a document is sensitive and you are not sure, stay with the on-device engine — it is the default for a reason.
Honest limits
Ten pages per run. A run reads the first 10 pages, and the result tells you plainly if the rest were skipped. Split a longer scan into parts of 10 pages or fewer with Split PDF and OCR each part.
Print, not handwriting. The engine reads printed and typed text. Handwritten notes come back garbled or empty — no setting fixes that.
No promise of perfection. Accuracy rides on the scan: a clean, straight page can read near-perfectly, and a rough one will have mistakes. The confidence score on every page is the honest measure — check low-scoring pages against the original before you rely on them.
A PDF that already has text does not need OCR. If you can already select and copy words from the file, its text layer exists — use PDF to Text or PDF to Word directly and skip the reading step.
About the tool
Make a scanned PDF searchable: recognise the text and add it as an invisible layer. Runs on your device; the first run downloads the language pack.
More about OCR PDF — what it does, formats and questions, or read this guide inside the app.
Questions & answers
Is OCR PDF really free?
Yes. The on-device engine is free with no sign-up, no trial period, no page charges and no watermark on either download, and you can run it as often as you like — one run reads up to 10 pages. The optional cloud engines are separate: they run on your own free key from those providers, and PdfLuxe charges nothing for choosing them.
Are my files uploaded anywhere?
Not with the default engine. Tesseract runs in your browser: your PDF is opened, read and rebuilt on your own device, and the results come straight back as downloads. The first run downloads the OCR engine and a language pack (a few MB) to your device — after that, nothing needs to travel at all. Only if you deliberately pick a cloud engine is each page image sent to that provider, on your own key.
How many pages can I OCR at once?
Up to 10 pages per run — a run reads the first 10 pages of the PDF, and if the document is longer the result message tells you the rest were skipped. For a long scan, split it into parts of 10 pages or fewer with Split PDF and OCR each part; each part gives you its own .txt and searchable PDF.
Which languages can it read?
Seven: English, Lithuanian, Hindi, Russian, German, Spanish and French. Pick the one printed on your pages under “Language on the page” before you run — the setting tells the engine which alphabet and word shapes to expect, and reading a page as the wrong language fills the result with wrong letters. The first run in a new language downloads that language's pack once.
How accurate is the result?
It depends on the scan, and any tool that promises 100% is not being straight with you. Each page is read three times and the clearest read is kept, and every page is reported with a confidence percentage; pages under 70% are flagged in the result so you know what to check. A straight, sharp, evenly lit page reads far better than a blurred or shadowed one — proofread anything that matters.
What do I get at the end — text or a PDF?
Both, your choice. The recognised words appear in the box on the right, page by page. Download .txt saves them as a plain text file; Download searchable PDF gives you the same pages as a PDF with the recognised text hidden inside as a real layer, so the file can be searched and copied from like any digital document — and converters such as PDF to Word can then read it.
Can it read handwriting, or a photo of a page?
Handwriting, no — the engine reads printed and typed text, and handwriting comes back garbled or empty. A photo, yes: under “Or read a photo directly”, pick a JPG or PNG photo of a page and its text is read on your device into the same box, and both downloads work — the searchable PDF puts the photo with its recognised text hidden behind it.
Why did a page come back empty or full of mistakes?
Usually the scan: a blurred, skewed, shadowed or very low-resolution page gives the engine too little to read, and a truly blank page honestly scores 0%. Check that the language setting matches the page, tick “Straighten a tilted page first” if the page leans, and re-scan the page flat, straight and well lit — a cleaner scan is worth more than any setting. A page from the middle of a long PDF may also simply be past the 10-page run limit.
Related tools
Related guides
This guide is one of 92 step-by-step guides on PdfLuxe — 151 free tools, no sign-up, no watermark. Questions? Email support@pdfluxe.com.