Is It Safe to Upload Client Documents to an OCR Tool?

It can be safe to upload client documents to an online OCR tool, but only after you have checked the vendor. You need to know what the tool keeps, who can see the file, whether uploads are used for anything beyond your conversion, and what the vendor has promised in writing. Without those answers, treat it as unsafe.

That is a diligence standard, and it does not guarantee anything. This post gives you the risk list, the questions, a fast way to read a privacy policy and the warning signs. It is practical guidance on handling documents, and it is not legal advice. Professional conduct rules differ from state to state, so check yours.

What are the real risks of uploading a client document?

The real risks are that the document is kept longer than you think, read by people you did not choose, reused for the vendor's own purposes, exposed in a breach, or left in a copy you forgot. Interception in transit is one more risk beside these.

An OCR upload differs from a chatbot prompt in one way that raises the stakes. You hand over the whole document, every page, including the parts you did not need converted. Go through the risks one at a time.

• Retention you did not agree to. The file, the extracted text and the export may each be stored, and each may follow a different clock. The post on whether OCR software stores your documents covers where copies sit and how to remove them.
• Human access. Support staff, engineers or contractors may be able to open files to fix problems or check for abuse.
• Secondary use. The vendor may use uploads to improve its product or train a model.
• Onward disclosure. The vendor may pass files to other companies that do part of the work, or disclose them in response to legal process without telling you.
• Breach. A service that keeps everything is a target, and you inherit its security.
• Account sprawl. A shared login, a departed employee's account, or a personal account used for firm work puts documents outside the firm's control.

Vendors are only part of the exposure. A phone photo in a synced camera roll and an export left in a Downloads folder are copies too. The section on where copies can linger lists them, and the guide to OCR on a phone photo of a document covers capture.

Which questions should you ask an OCR vendor?

Ask what is stored and for how long, who can open your files, whether uploads feed any model or product improvement, which other companies receive the data, what happens after a breach or a subpoena, and how deletion works. Ask for the answers in writing, and for the document that makes them binding.

Bar guidance supplies the outline. ABA Formal Opinion 512, drawing on earlier opinions about cloud computing and outsourcing, suggests that lawyers confirm a tool is configured to preserve confidentiality and that "the obligation is enforceable", that they will be notified of "a breach or service of process", and whether the provider "retains information submitted by the lawyer before and after the discontinuation of services or asserts proprietary rights to the information." Opinion 512 is about generative AI, and many OCR products now run on AI models, so its list fits. The ABA's opinions are advisory, and your own bar may ask for more or less.

Two follow-ups separate a careful vendor from a careless one. First, ask whether the answers differ by plan. Free tiers can run under different terms from paid ones. Second, ask what would change the answers, and how you would find out.

Send the questions by email and file the reply with the matter or with the firm's vendor records. A written answer is evidence of the reasonable efforts your rules are likely to require. A phone call leaves you with nothing.

File that reply where the next person on the matter can find it. Choosing a converter is part of the duty of competence, and client confidentiality is what the choice protects. Information security questions belong in the same email as the retention questions. Ask about third-party risk by name: which other companies touch a page, and on what terms.

Reading a privacy policy in five minutes

You can triage a privacy policy in five minutes by searching it for a short list of words. You are not reading for style. You are looking for five facts: what is collected, what it is used for, who receives it, how long it is kept, and how deletion works.

The State Bar of California's practical guidance on generative AI says a lawyer should take reasonable steps to understand how a product "collects, uses, stores, and discloses information", and that reasonable efforts "require more than reliance on generalized marketing assurances." So skip the homepage and open the policy.

1. Find the right documents. Open the privacy policy and the terms of service. Check the footer for a separate data processing agreement or security page. Note the "last updated" date on each.
2. Search "content", "upload", "files" and "documents". Many policies talk only about account data such as your email address and say little about the files themselves. If uploaded content is not addressed, note it as an open question for the vendor.
3. Search "improve", "train", "develop", "research" and "machine learning". These words mark secondary use. Read each whole sentence.
4. Search "retain", "retention", "delete" and "backup". Look for a number of days. "As long as necessary" is not a number.
5. Search "share", "disclose", "third part", "service provider", "affiliate" and "subprocessor". Write down every category of recipient.
6. Search "aggregate", "de-identif" and "anonymi". These often introduce an exception to a promise made elsewhere.
7. Check the terms for a licence. Search "license" and "licence". A clause granting the vendor a broad right to use your content is a different promise from the privacy page.

When five minutes is up, write three lines: what the policy says on retention, on secondary use and on disclosure. Add anything it left unanswered. Save the policy as a PDF with the date. If the terms change later, you will want the edition you relied on.

A five-minute read is triage. For a tool the whole firm will use, someone should read the terms in full. Opinion 512 says a lawyer may instead consult a colleague or outside expert who has read and analyzed them.

Signs that a converter is not safe for client files

A converter is not safe for client files when you cannot tell who runs it, what it keeps, or what it does with uploads. Vague terms, no way to delete, and the absence of any account or contract are the clearest signs. Any one of them is reason enough to pick something else.

• No company name, address or contact beyond a web form.
• No privacy policy, or one that never mentions uploaded files.
• Terms that grant the operator a licence to use, reproduce or publish your content.
• "We may use your content to improve our services" with no way to refuse.
• No stated retention period and no delete control.
• No account at all. With no account, nobody has made you an enforceable promise, and you cannot go back and delete.
• Public or guessable links to your converted file.
• A page crowded with advertising and trackers around the upload box.
• A policy that changes "at any time without notice".
• No answer when you send the questions from the table in this post.

Free tools are not all alike, and some paid services offer free tiers on proper terms. The post on whether free online OCR is safe for legal documents sorts out which is which.

No vendor at all is the right answer for some documents. If a protective order names who may receive designated material, or a client has told you a file must not leave the firm's systems, the vendor's policy does not come into it.

What Schema Law commits to

Schema Law lists six lines that apply on every plan, four of them about how documents are handled. They are published on its pricing page and described under security on the homepage. They are the company's own statements. Quote them, ask about them, and weigh them the way you would weigh any vendor's.

The four, as written:

• "Your documents are never used to train an AI model"
• "Documents stay inside your own account"
• "Encrypted in transit and at rest"
• "Delete any document, at any time"

The homepage expands on the second: "Documents stay inside the account that uploaded them. Nothing is pooled across firms, published, or made visible to other customers." No plan has a retention setting. Nothing is deleted automatically, so a document stays until you delete it. On deletion the homepage says: "Delete any document whenever you choose. The original file and everything parsed from it are removed."

A list of commitments is where diligence starts. Nothing in this post independently verifies those statements, and you should not read it as doing so. Put every question in the table above to Schema Law, and ask which document makes each commitment binding on your plan. The Enterprise plan lists "Security questionnaire and contract review" as a feature, for firms with their own security review. On any other plan, use the contact address on the site and keep the reply.

Two related posts go further on single commitments: how to verify a no-training claim and what encryption does not cover.

Client consent and engagement letters

Whether you need a client's consent before using an OCR service depends on your jurisdiction's rules, the tool, and what the client has told you. Where a tool creates a real risk of disclosure, bar guidance points toward specific, informed consent.

Start with the rule. ABA Model Rule 1.6(a) bars revealing information relating to the representation without the client's informed consent, implied authorization or a listed exception. The section on what the confidentiality rules require of a lawyer quotes it. Your state's version may be worded differently. How far implied authorization reaches for a given vendor is a question for your bar's opinions, and this post does not answer it.

Opinion 512 addresses one end of the range. For self-learning generative AI tools, it says a client's informed consent is required before the lawyer inputs information relating to the representation, and that "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient." It also calls the engagement agreement "a logical place" to disclose how the firm uses such tools and to record any client instructions.

In practice:

• Read what the client already told you. Check outside counsel guidelines and engagement terms for limits on vendors or a duty to give notice.
• Describe the practice plainly. If you address document conversion in an engagement letter, say what kind of service is used, for what, and what safeguards you checked. A sentence the client can understand does more than a paragraph of boilerplate.
• Offer a way to say no. A client with an unusually sensitive matter may want certain documents kept off outside services.
• Go back to the client when something changes. A new tool, or new terms from an old one, can change the analysis.

A short worked example. A family law client hands over five years of bank statements to be converted to spreadsheets. The firm's engagement letter says it uses a document conversion service it has vetted, names the safeguards it confirmed, and invites the client to ask for any document to be handled manually. The client's file notes show the vendor's written answers. Your rules may require more than that or less, but with a record like this you can show what you decided and why.