I spent a day reading the privacy policies of ten free plagiarism checkers, looking for one answer: after the scan finishes, what happens to your paper? Two of them say your text may be used to train their AI models unless you email and ask them to stop. Three of them publish the same privacy policy, word for word, as each other. One never mentions your uploaded document anywhere in the document.

Only a few gave a straight answer. Below is what each one actually says, and how to run this check yourself on any tool in about two minutes.

All policies were read on July 23, 2026. Policies change, sometimes quietly, so treat this as a snapshot and a method rather than a permanent scorecard.

What I was looking for

Three questions, in order of how much they should worry you:

Is your text used to train the company's models? Once your draft is training data, it is not coming back out.
Is your text added to a matching database? This is the one students fear most, because a paper indexed today can match against your own submission tomorrow.
How long is your text kept, in actual units of time? "As long as necessary" is a purpose, not a period.

Two checkers train on your text by default

This was the finding I did not expect to be so explicit.

Quetext's policy has an entire section on AI training, evaluation, and data licensing. By submitting content you agree that it may be included in the company's model training, fine-tuning, and evaluation. There is an opt-out, but it works by emailing support with a specific subject line, it applies only going forward, and the policy is candid that models already trained cannot practically be untrained. The same section reserves the right to license aggregated and derived data commercially to third parties, and says aggregated and derived data may be kept indefinitely.

Copyleaks lists the materials you upload, your texts and document scans, among the user data it collects, and states that this information is used to train its models. Direct customers can opt out by contacting support by email. Users who come in through an institution have to rely on that institution to opt out on their behalf.

Neither of these is hidden or unlawful, and both disclose it more plainly than most companies would. But the default in both cases is participation, and the exit in both cases is an email you have to know to send. A student pasting an essay into a free checker at 1 a.m. has not read section six.

Three different sites, one identical policy

Duplichecker, Prepostseo, and SmallSEOTools present themselves as separate products. Their privacy policies are the same text. Same definition of user content, same sentence about log files opening with "Same like other websites," same promise that employees do not monitor all user content, same account-deletion section.

Duplichecker's copy carries one clause worth reading twice: it lists, among the reasons staff may view your uploaded content, a need to review it "to enhance our algorithms" as described in its terms of service. Whatever that means in practice, it is not the same as nobody ever looking at your draft.

All three describe deletion in terms of your account: cancel a premium subscription, delete the account, and your personal data goes. None of them commits to a deletion timeline for the text you pasted in as an anonymous visitor, which is how most students use these sites.

The ones that answer the actual question

Scribbr was the clearest of the ten. Its help documentation states directly that the self-plagiarism checker does not store your document in any public database, and that you can delete your personal information and documents from its servers once you have your report. That is a real answer to a real question, in language a stressed student can act on.

Worth knowing: PaperRater and Scribbr both now run under the same Learneo privacy policy, which is being updated on August 21, 2026. So when you compare those two, you are not comparing two independent companies' privacy practices.

"As long as necessary" is not a retention period

Grammarly collects the content you upload and says it retains personal data as long as needed to provide its products, meet legal obligations, and resolve disputes, with different data kept for different periods depending on how you configure your settings. The Learneo policy covering PaperRater does something similar and adds that storage periods are worked out case by case.

Both statements are standard, defensible, and almost certainly written by careful lawyers. Neither tells you when your essay is deleted. That is the gap: the language is legally sound and practically useless to the person deciding whether to paste a thesis chapter into a text box.

The policy that never mentions your document

Check-plagiarism.com publishes what looks like a standard generated privacy template. It covers cookies, usage data, retention of personal data, and your rights in detail. Searching it for anything about the documents users submit to the plagiarism checker turns up nothing specific.

Silence is not evidence of bad behavior. It is evidence of nothing at all, which is the problem: you cannot rely on a commitment a company never made. If a checker's policy does not mention submitted content, you have no basis to assume it is deleted.

Where we land, including the caveat

Our turn. It would be cheap to run this audit on nine competitors and skip our own policy, so here it is. Our checker promises zero retention on scanned text, which means it sits in memory while the scan runs and then it is gone. No indexing. Nothing added to a matching database, nothing sold. If you never sign in, all we keep is a salted hash of your IP address rather than the address itself.

The caveat, since a comparison that hides its own exception is worthless: if you are signed in and choose to save a report to your check history, that report is stored for your account until you delete it. That is a feature you opt into, not a default, and the dashboard lets you delete individual reports, your whole history, or the account. Our other honest limit is the 2,000-word cap per scan, which means a full dissertation goes through in sections.

How to check any checker in two minutes

You do not need to read a policy end to end. Open it and use your browser's find function on these terms:

"train": catches model-training clauses, which are the ones with permanent consequences.
"user content" or "submitted content": find how the company defines your text, then read what it says it may do with that category.
"retention" and "delete": look for a number. Days, hours, "immediately after the scan." No number means no commitment.
"database" or "index": confirms whether your paper becomes something future scans match against.
"third part": catches both sharing and licensing clauses.

If a search for your own document turns up nothing, that is your answer for that tool. And if you are pasting in work that is not yours to share, a client's confidential draft, an unpublished manuscript, a paper with someone else's personal data in it, do this check before you paste, not after.

What this actually changes

For a first-year essay, most of this is low stakes. For a dissertation chapter you will publish, an unpublished paper under review, or client work under an NDA, the difference between a tool that deletes your text after the scan and one that trains on it by default is a difference that matters, and it is knowable in two minutes.

The pattern across all ten was consistent: the tools that answer the question answer it plainly and near the top, and the tools that do not answer it use many words to not answer it. That holds well beyond plagiarism checkers. It is also worth remembering that a checker's privacy policy is a separate question from what your institution does with your work, since the paper you submit through your university's system follows your school's rules, not the tool's.