OCR Text Recognition Online: What Happens After Upload

OCR text recognition online is recognition you reach through a web page. You pick a scan in your browser, the page passes it to recognition software, and text comes back to the same page. The difference from installed software is where that reading happens: on the provider's computers or, with some tools, inside your own browser.

For a lawyer, "where" is the whole question. The recognition itself is the same technique either way. What changes is whether a copy of the client's document has left your machine, and what the provider's site says happens to it next.

What does OCR text recognition online mean?

OCR text recognition online means recognition delivered as a website instead of a program you install. You need a browser and a connection. You do not need a licence key, an installer or a particular operating system. Searches for an "OCR scanner online" or a way to "scan text online" are looking for this kind of tool.

The names vary and the thing does not. "PDF text recognition online", "OCR text scanner online" and "online OCR" all describe the same arrangement: a page with an upload control, recognition somewhere behind it, and a result to copy or download.

Two points of vocabulary save confusion.

Nothing is scanned. An online "scanner" has no glass and no lamp. It starts from a file that already exists: a PDF from the office copier, a photo from a phone, a fax saved as an image. The post on the OCR text scanner separates the device that makes the picture from the software that reads it.

Online describes delivery, not quality. The same recognition core can sit behind a website, inside a desktop program or in a copier. Adobe, for instance, describes its browser tool on the page for its free online OCR tool this way: "You can use Adobe Acrobat tools in any browser, like Microsoft Edge or Google Chrome, so there’s no additional software you need to install". Whether a browser tool or installed software suits your firm is a separate decision, set out in the comparison of online OCR and desktop OCR for law firms.

Files that are not scans

Online tools are built around pictures of pages: scanned PDFs and image files. A Word document or slide deck with a screenshot pasted into it also holds a picture of text, and the words in that picture are invisible to search for the same reason. That case is covered in the post on whether you can OCR a Word document.

What happens to a page between upload and recognized text?

Between upload and result, a page goes through four steps: the file is transferred, it waits its turn, recognition runs page by page, and the text is sent back to your screen. With a server-based tool all four happen on the provider's side. You see a progress bar and then a result.

Here is the sequence in more detail, for a tool that does the reading on a server.

1. Selection. You choose a file or drag it onto the page. The browser now holds the file in memory but nothing has been sent.
2. Checks in the browser. The page may test the file's type and size and refuse it at once if it breaks a limit.
3. Transfer. The browser sends the file to the provider. The connection is encrypted when the page's address begins with https. On a slow line this can be the longest wait, and it grows with megabytes, not pages.
4. Checks on the server. The service confirms the file is what it claims to be and counts its pages against your allowance.
5. Queue. If other jobs are ahead of yours, the file waits.
6. Page preparation. A PDF is split into pages, and each page is turned into an image the reader can work on.
7. Recognition. Each page image is read into text. The steps inside this stage are explained in the post on how OCR works.
8. Assembly. The text for each page is gathered, with whatever structure the tool keeps, into a result.
9. Return. The result is shown in the browser or offered as a download.
10. Afterlife. The uploaded file and the result are kept, or deleted, according to the provider's rules.

Adobe's page compresses steps 3 to 9 into two sentences: "After the file uploads, the tool will automatically apply Optical Character Recognition (OCR) to detect and convert text from images or scanned pages. Sign in to download your newly searchable PDF or get a link to share it."

What you can and cannot see

From your chair, steps 1, 2, 3 and 9 are visible. Steps 4 to 8 happen out of sight, and step 10 is invisible unless the site tells you about it. That is the practical meaning of "online": the part of the process that matters most to a confidentiality analysis is the part you take on the provider's word.

When it fails partway

A failure at each step looks different, which helps when a deadline is close.

Does online text recognition run in your browser or on a server?

Online text recognition can run in either place. With a server-based tool the file is uploaded and read on the provider's servers. Some tools instead download the recognition software into your browser and read the page on your own computer. The tool's own description should say which, and you can check by watching what the page sends.

Reading on a server

This is the arrangement the word "upload" implies. A copy of your file travels to the provider, is read there, and the text travels back. Schema Law's homepage describes an upload "straight from your browser", and one of its commitments reads: "Files are encrypted from the moment they leave your computer and stay encrypted while stored." The full list is under security on the homepage.

Reading in the browser

A browser can run recognition code itself. The open-source library Tesseract.js is one way a site can do it. Its project README says the library "works in the browser" and on a server, and describes its aim as bringing the Tesseract engine to the browser "by wrapping a WebAssembly port of Tesseract." A site built this way sends the recognition program to you, and your computer does the reading.

One detail from the same README is worth knowing before you assume a browser-side tool will handle a legal file. It states in bold that "Tesseract.js does not support PDF files", so a site built on it needs other code to handle a scanned PDF. A browser-side tool also still downloads something, namely the recognition program it runs.

A tool that reads in the browser may still send your file, or the resulting text, to a server for other reasons, such as saving it to an account. In-browser recognition and "nothing leaves your computer" are two separate claims. Look for the second one in writing.

How to tell which you have

1. Read the site. Look for a plain sentence about where files are processed. "Processed on our servers" and "processed locally in your browser" are both clear. Silence is an answer of a kind.
2. Watch the upload. A progress bar that crawls on a large file suggests a transfer. An instant start on a 15 MB scan suggests the file stayed put.
3. Look at the network traffic. Browsers include a developer tool for this. Firefox's documentation says its Network Monitor "shows you all the HTTP requests Firefox makes", with details of each request. Open it, upload a test page that is not confidential, and look for an outgoing request about the size of your file.
4. Disconnect and try. Load the page, turn off the network, and choose a harmless test file. A server-based tool cannot proceed. A browser-based one may, if it has already downloaded what it needs.
5. Ask. If the answer matters to the matter, put the question to the provider in writing and keep the reply.

Use a page with nothing sensitive on it for every one of these tests.

Why online recognition tools cap pages, megabytes and file types

Online tools set limits because every upload uses the provider's bandwidth, storage and computing time, and because a file type the software cannot open is better refused at the door than failed halfway. The caps on megabytes, pages and formats are separate limits, and a document can pass one and fail another.

Size and pages are different things

Megabytes measure the pictures, and pages measure the work. A two-page color scan of a site plan can be larger than a hundred-page black and white deposition transcript. Check both figures for the file in front of you before you plan around either.

Limits as two providers state them

Google's help page on how to convert PDF and photo files to text in Google Drive lists its preparation tips as a short specification. It names the formats, "PDFs (multipage documents) or photo files (.jpeg, .png and .gif)", and says: "The file should be 2 MB or smaller."

Schema Law's figures are on its site. The homepage says an upload is "A PDF, PNG, or JPEG of up to 20 MB". The plans then set page limits:

The pricing page has the current terms. The two providers' numbers are quoted here to show the shape of a limit, not to rank anyone. A size cap tells you nothing about how well a tool reads.

Working within a cap

• Too many megabytes. Rescan in grayscale or black and white if color is not needed, or split the file.
• Too many pages. Split at natural breaks, such as exhibits or volumes, and name the parts so they can be reassembled in order.
• Wrong type. Convert to PDF with the software that made the file, and check that the conversion kept every page.

Splitting has a cost in legal work. Page numbers in the output restart with each part, so record the original page range of every part before you upload it.

What free online text recognition usually leaves out

"Free" describes the price of trying the tool. It does not describe what the tool does with your file, how much you can convert, which formats you get back, or whether an account is required. Each of those is a separate term, and for a search such as "pdf text recognition free" they matter more than the price.

Check five things, and treat each as a question about the specific tool, not as a rule about free tools in general.

Adobe's page illustrates the second and third rows in its own words. It says "The Acrobat OCR online tool lets you recognize text in a PDF document for free", and that to work with "other documents, including image file formats like PNG, JPG, and TIFF files, you can start a seven-day free trial of Adobe Acrobat Pro." It also says to "Sign in to download your newly searchable PDF". None of that is hidden. It is simply what the word free covers on that page.

Schema Law's free plan is stated the same way, as a list: one document a day and up to three pages per document. No export format is held back: every plan, including the free one, exports Word, Excel, CSV, JSON, searchable PDF, plain text, Markdown and HTML, and accepts PDF, PNG or JPEG uploads of up to 20 MB, several at a time. The commitments that apply to every plan are that documents are never used to train an AI model, that they stay inside your own account, that they are encrypted in transit and at rest, and that you can delete any document at any time.

What a free tier is good for

A free allowance is a way to test a tool on your own pages before you rely on it. Three pages are enough to learn a good deal if you choose them well: one clean typed page, one poor fax, and one page with a table or figures. Read the names, dates and amounts in the result against the originals.

A free allowance is the wrong place for a client's confidential document until you have read the terms. The post on whether it is safe to upload client documents to an OCR tool lists the questions to settle first.

Reading an online OCR tool's own description before the first upload

Before the first upload, an online OCR tool's site should tell you who runs it, where files are read, what is kept and for how long, whether content is used for anything beyond your conversion, what the limits are, and how to delete your documents. If you cannot find those answers, do not send it a client file.

Work through the site with a list before the tool becomes a habit.

What to look for, and where

Reading the answers

Prefer a sentence to an adjective. "Secure" and "private" are not commitments. "Delete any document whenever you choose. The original file and everything parsed from it are removed" is one, and it is how Schema Law's homepage puts it.

Notice what is absent. A site that says nothing about retention has not promised to delete anything.

Look for candor about quality. A tool that claims to read everything perfectly has told you nothing. Schema Law's answer to "How bad can the scan be?", under questions partners ask first, says recognition is strongest on clean print and weaker on faint faxes, stamps over text and handwriting, and that nothing marks a passage as uncertain. You may want more than that from a tool, but you know where you stand.

Match the description to your duties. Your obligations about client information come from your jurisdiction's rules and your engagement terms, not from the vendor. The site's description is the input to that analysis, so save a dated copy of the pages you relied on.

Test with a harmless page first. Upload something public, such as a page from a published court form, and walk through the whole cycle: upload, result, export, delete. Confirm the document is gone from your list afterwards.

Try a single transcript page before a whole production. Search the returned text for one party name and one citation you can see on the sheet. If those strings disagree with the ink, stop. You would carry a wrong fact onto every later exhibit in the batch.

Only then send the file that matters.