Private OCR: What Makes a Converter Confidential

Private OCR means a converter reads your document without anyone outside your control seeing it, keeping it, or learning from it. An OCR tool earns the word in one of three ways: it runs on your own computer, on servers your firm controls, or as a closed service with written limits on access.

"Private" is a marketing word with no fixed meaning, so two converters that both claim it can treat the same lease very differently. This post is not legal advice, and the duty you owe a client is set by the ethics rules of your jurisdiction.

What does private OCR actually mean?

Private OCR means you can answer five questions about a document and like every answer: who can see it, where copies exist, how long they last, whether it is reused, and which colleagues can open it. A converter that leaves any of the five unanswered is asking you to take its privacy on trust.

Three words get used as if they were the same, and they are not:

• Secure OCR is about outsiders. It describes defences against interception and theft, mainly encryption and account security. The post on encrypted OCR covers what those defences do.
• Private OCR is about the provider and everyone it works with. It describes what the people running the tool may do with your file.
• Confidential describes your obligation to the client. A tool can support it. Only you can meet it.

A converter can be secure without being private. A service can encrypt every upload and still keep the right to store files and study them to improve its product.

You still owe client confidentiality after you pick a route, and the duty of competence is being able to say who can open the file. Information security covers the laptop and the account together. Write the route in the matter note so the next person on the file follows it instead of improvising. If a protective order names permitted recipients, check that list before the route is treated as settled.

The fourth row is the one that newer, model-based converters have made urgent. The post on OCR that does not train on your data explains how to test that promise.

Three ways to keep OCR private

You can keep OCR private by running it on your own computer, by running it on servers your firm operates, or by using a closed service that commits in writing to what it will and will not do. Each moves the trust to a different place. None removes the need for it.

Local software. The program is installed on your machine and the page is recognized there. Commercial desktop suites work this way, and so do free engines. Tesseract, for one, is described in its own documentation as "an open source text recognition (OCR) Engine" that "doesn't have a built-in GUI", which means someone has to be comfortable with a command line or install a separate front end.

Self-hosted. The firm runs OCR software on a server it controls, and staff reach it over the office network. Documents never go to an outside provider, and everyone shares one setup. The cost is that the firm now owns installation, updates, backups and access control.

A closed service. Another company runs the software and you upload through a browser. Privacy comes from the provider's commitments: your files are kept apart from other customers' files and are not reused. The post on closed system legal AI sets out what that term should cover and how to test it.

The comparison of features, cost and upkeep between installed and browser tools belongs to the post on online OCR versus desktop OCR for law firms. This table is only about who can reach the document.

Is offline OCR always safer?

No. Offline OCR removes the provider from the list of people who could see a document, and that is a real gain. It leaves every risk that lives on the computer itself, and it adds a few of its own. Offline is safer only when the device is well kept.

Consider what a laptop is exposed to. NIST's guide to storage encryption for end user devices, Special Publication 800-111, names unauthorized access to information on a lost or stolen device as the main threat that storage encryption addresses. It also warns that once a user has authenticated, "malware could access decrypted files and transfer copies of them to external hosts". A converter that never touches the internet does not help if the machine it runs on is compromised or left in a taxi.

Three ways local conversion leaks in practice:

1. Synced folders. The scan and the Word export are saved to a desktop or documents folder that a consumer cloud-storage app copies off the machine automatically. The OCR was offline. The files were not.
2. Cloud features inside desktop software. Some installed programs offer optional online functions. Check the settings to see whether any recognition, storage or assistant feature sends content out, and read the vendor's documentation for the version you run.
3. Leftover copies. Scans sit in the copier's memory, in the email that carried them from the copier, and in the downloads folder. Private conversion does not clean up the route the document took to reach you.

There is also a quieter risk. If the local tool handles a faxed or crooked page badly, staff under deadline will look for something that works, and the nearest thing is a free site. A private tool that people avoid protects nothing. Test any local option on your worst real scans before you make it the only permitted route.

So the honest comparison is between a device you manage and a provider you have vetted. For a firm with managed, encrypted laptops and an IT person, local is a strong choice. For a firm where each lawyer looks after their own machine, a vetted closed service may expose fewer copies to fewer hazards.

Private PDF to Word conversion in practice

A private PDF to Word conversion keeps count of every copy made along the way and removes the ones you do not need. The conversion is the easy part. The copies in the scanner, the inbox, the downloads folder and the converter are where a confidential file usually ends up exposed.

Take a 14-page signed settlement agreement that arrives on paper. You need it in Word to draft an amendment. Here is the same job by each route:

Count the copies at the end. You should be able to name each one: the paper original, the scan in the matter folder, and the Word file beside it. Anything else is a stray.

Two mistakes turn a confidential PDF to Word conversion into a leak. The first is emailing the scan to yourself to move it between devices, which leaves a copy in two mailboxes. The second is leaving the file in the converter for months after the work is done.

In Schema Law the route is the one in the right-hand column: sign in with Google, upload the PDF, export Word, and delete the document when you are finished. Every plan lists "Documents stay inside your own account" and "Delete any document, at any time" among its commitments, and the security section of the homepage gives the wording in full. Those are the company's statements about itself. Ask it the five questions from the first section, in writing, as you would ask any other provider.

For the conversion steps themselves, and what to check in the Word file, see the guide on how to convert a scanned PDF to Word.

Who else can see a document in a shared workspace

In a shared workspace, everyone the firm has added may be able to open a document unless access is limited by matter or by person. Privacy from the outside world is one question. Privacy from colleagues who are screened from the matter, or who have left the firm, is another, and tools differ widely on it.

This matters most where the firm has put up an ethical wall. If a lateral hire is screened from a matter and the OCR workspace shows every upload to every user, the screen has a hole in it.

Ask these of any shared tool:

• Can a document or folder be limited to named people, or does every user see everything?
• Can one person be shut out of one matter?
• What can an administrator see?
• When someone leaves the firm, how is their access removed, and does it happen the same day?
• Do several people share one login? If so, nobody can tell who opened what.
• Can the provider's support staff open documents, and under what conditions?

Schema Law has no shared workspaces, so most of that list does not arise. Each account is one person's Google sign-in, and the documents that person uploads sit in that account. Every plan lists "Documents stay inside your own account", and no plan adds colleagues, matters or a firm-wide view. The fifth question still applies. If several people at the firm use one Google sign-in, each of them sees everything uploaded under it, so give each person their own account. Pass exports to colleagues through the firm's own document system, where your screens between matters already apply, and put the last question on the list to Schema Law in writing. The pricing page shows the plans side by side.

Matching the level of privacy to the document

Match the route to the document by sorting what you hold into a few classes and deciding once what each class requires. Most paper in a law office needs a vetted tool and sensible cleanup. A small share needs more, and a small share needs nothing special.

Two cautions apply to that table. A protective order or a client's outside counsel guidelines may name who can handle a document, and that wording overrides any general rule the firm has made. And the standard is yours to meet. The D.C. Bar's Ethics Opinion 388, writing about generative AI, says a lawyer "should be reasonably satisfied" that using a tool will not reveal client confidential information to its future users. That opinion speaks for one jurisdiction and one kind of tool. Check what your own bar requires.

When a document sits between two classes, treat it as the higher one. Then write the classes down so that nobody has to work them out again on a deadline. The guide to a law firm AI policy for document uploads shows how to turn the classes into a register of permitted tools.