Do online PDF tools keep your files? I read iLovePDF, Smallpdf, Sejda, and Adobe's policies
I read the privacy policies of four popular PDF tools to find out how long they keep your file, who can reach it, and whether it can train AI. Here's what each one says, quoted.
Most people drop a contract or a bank statement into an online PDF tool without thinking about where it goes. I built a browser-only alternative, so I have an obvious bias here. To keep it fair, I stuck to what the companies themselves wrote. On October 7, 2026 I read the current privacy policies of four popular services and looked for three answers: how long they keep your file, who else can reach it, and whether it can be used to train AI.
Policies change, so treat everything below as a dated snapshot, not legal advice. Every quote links to the page it came from.
iLovePDF
- Retention:files are deleted “within TWO (2) HOURS of being processed on ILOVEPDF's servers” (iLoveSign documents follow their own terms).
- AI training:the policy says, in capitals, “ILOVEPDF DOES NOT USE YOUR CONTENT TO TRAIN, IMPROVE, OR DEVELOP ITS AI MODELS OR ANY THIRD-PARTY AI MODELS.”
- Where:processing happens in the European Economic Area by default, but for users outside it the company “may process your personal data in geographic areas closer to your location.”
This is the clearest of the four. A fixed two-hour window and a flat no on AI training are easy to hold a company to. Source: iLovePDF privacy policy.
Smallpdf
- Retention:if you're logged in, files are deleted “within one hour unless you save them.” Without an account, Smallpdf aims to delete them “within a reasonable period of time after the last time they were opened,” and that period “is extended every time you reopen” the file. No number of hours is given.
- Sharing:“By default, the documents you process using our services are accessible to anyone with a unique sharable URL. This can be disabled by the User.”
- AI training:for its AI features (chat, summarize, translate), “We do not use these files to train models.”
The shareable-by-default link was the line that surprised me most. The link is unique, so nobody stumbles onto it by accident, but anyone you forward it to, or anyone who finds it in an email, can open the document until it's deleted.
One sentence will probably alarm people, so it's worth reading in context. In the section on your privacy rights, the policy says “We currently sell data to Google and Facebook via cookies.” That refers to advertising cookies on the website, which count as a “sale” under California's privacy law, and you can opt out. It is not about the PDFs you upload. Source: Smallpdf privacy notice.
Sejda
- Retention:files are kept “for the sole purpose of giving you enough time to process and download them” and then permanently deleted. “No backups are made of these files.” The policy doesn't give an exact number of hours.
- Sharing:only if you create a public link yourself. Shared files are deleted “7 days after being shared.”
- AI training:AI-assisted features send content to a sub-processor (listed as Google Cloud) for processing that “is transient, and your files are not used to train AI models.”
Sejda is unusually specific about what happens with AI features, including where they run (the EU, with the US as failover). Source: Sejda privacy policy.
Adobe Acrobat
- Generative AI:“We do not analyze your content to train generative AI models, unless you choose to submit content to the Adobe Stock marketplace.”
- Other analysis:Adobe describes a separate “content analysis for product improvement” covering content processed or stored on its servers, and says you're in control of it through an opt-out.
- Retention:I didn't find a fixed deletion window for files processed by Acrobat's online tools in the pages I read.
Adobe draws a line between training generative AI, which it rules out, and analysing content to improve products, which it allows unless you opt out. If that distinction matters to you, check the setting in your account. Source: Adobe content analysis FAQ.
What stood out
All four say, in writing, that your uploaded files don't train generative AI. That's good news, and it's the question I hear most often. The real differences are in the quieter details: how long the file sits on a server, whether a link to it exists by default, and whether there's a clear number or just “a reasonable period.”
Every policy also includes the usual exceptions: hosting providers that process files on the company's behalf, and disclosure when the law requires it. That's normal, but it means “we delete your file” is a promise about the company's own behaviour, made under conditions you can't check from the outside.
A policy is a promise, not a guarantee
None of this says these services are unsafe. It says that once a file leaves your device, the only protection is the company's word and its security. For most documents that's fine. For a contract full of personal data, a medical record, or anything covered by an NDA, the simplest answer is not to upload it at all.
That's why PDFShore's tools run in your browser instead of on a server. They can't do everything a cloud service can (there are no shared links or e-signature audit trails, for example), but for compressing, merging, redacting, or counting tokens, the file never leaves the tab. You don't have to take my word for it either: here's how to check whether any PDF tool uploads your file, mine included.