Smallpdf, iLovePDF, PDF24 and the rest are competent products. They are fast, they handle formats a browser cannot, and their privacy policies are clearer than most. Nothing below is an accusation.
But there is a structural fact underneath all of them, and it is the reason the “alternative” searches exist: to merge your PDF, they need your PDF. It goes to their servers first.
The two architectures
Server-side. Your file is transmitted over the network, written to disk or object storage, picked up by a worker process that runs the operation, and written out again. You get a link to the result. Along the way the document exists on their web server, in their storage, in whatever queue moved the job, in backup snapshots, and in logs.
Reputable providers delete it after a window — often an hour — and say so in writing. That promise is usually kept. It is still a promise about how a company behaves, and it depends on their retention policy, their staff, their subprocessors, their jurisdiction, and their security holding.
Browser-side. The page loads code, that code reads the file you picked with the standard File API, and the operation runs on your own processor. The result is handed back as a download. Nothing is transmitted, because there is no endpoint to transmit to — the site is static files on a CDN.
The difference is not that one company is more trustworthy than another. It is that one arrangement asks you to trust a policy and the other removes the thing the policy is about.
Where server-side genuinely wins
This matters, and most comparison pages skip it because it weakens the pitch. Three things a browser cannot do well, or at all:
OCR. This one is a choice rather than a limit, and it is worth saying so plainly: OCR does run in a browser — tesseract.js is a real library and the English model is about a megabyte. What it is not is as good. On a clean scan it is fine; on a skewed phone photograph of a receipt it is well behind what a modern cloud service manages, and it reads Arabic worse than it reads English. Each language is its own download. We would rather point you at something that does it properly than ship a weaker one and let the result be your problem.
Serious compression. Genuinely shrinking a PDF means re-encoding the images inside it, which is what Ghostscript and its commercial equivalents are for. A browser can strip metadata and rewrite the structure; it will not match those results.
PDF to Word. A PDF stores glyphs at coordinates. It does not know where a paragraph starts or what a table is. Reconstructing an editable document is inference, and the good implementations are large server-side engines.
And very large files. Browser tools hold the document in your device's memory. A 500 MB scan will defeat a phone and may defeat a laptop. A server has no such limit.
If you need any of those, use a server tool and choose one with a retention policy you have actually read. That is the honest answer.
How to check which kind you are using
You do not have to take anyone's word for this, including ours. Two tests, both under a minute, and they work on any site.
Watch the network. Open your browser's developer tools (F12), switch to the Network tab, then use the tool. If your file is being uploaded you will see a request carrying it — a POST of roughly the size of your document. If the work is happening locally, no such request appears.
Switch off your connection. Load the tool page, then turn off your Wi-Fi and use it anyway. A browser-side tool keeps working. A server-side one cannot, and will tell you so.
The second test is the more useful of the two, because it needs no technical knowledge and it cannot be faked. A tool that works with the network switched off did the work on your machine. There is no other explanation.
What this site is
Termiva is the browser-side kind: 88 tools, no accounts, no upload, no daily limit. Merging, splitting, rotating, cropping, watermarking, page numbers, images to PDF and back, metadata removal — all of it on your own device.
It does not do OCR, PDF to Word, or real image-level compression, for the reasons above. Those are left out rather than done badly.
The claim is checkable with either test on this page, which is the point of making it this way round: not “we delete your files quickly” but “we never receive them”.
The short version
If your document is a takeaway menu, none of this matters and any tool will do. If it has your name, signature, salary or account number on it, the question worth asking is not whose privacy policy you prefer — it is whether a policy needs to exist at all.