Why your scan behaves like a photograph
Because it is one. A scanner takes a picture of the paper, and a PDF is perfectly happy holding a picture. What you are looking at is an image of words, and there is no text in the file at all.
Which explains every symptom at once. Copy and paste gives you nothing. Searching finds nothing, even for a word you can plainly see. Converting it to Word produces an empty document. None of that is a fault in the tools; there is genuinely nothing there for them to find.
Optical character recognition is the only fix. It looks at the shapes in the picture, works out which letters they are, and writes real text back.
How this tool works
Worth knowing, because it explains both what you get and why it takes a moment.
Each page is drawn at 300dpi. Recognition engines are tuned around that resolution: much lower and small print falls apart, much higher is slower without reading any better.
The picture is read. The engine is Tesseract, which is what almost everything in this field uses, compiled to run in the browser. It returns every word it found, how confident it was, and exactly where on the page it was.
The words go back into the original. Each one is drawn at the position it was found, at zero opacity. It paints nothing, so the page looks exactly as it did, and it is real text in the file, so readers search it, select it and copy it. That is all a searchable PDF is.
The original pages are left completely alone. Nothing is rasterised, re-encoded or compressed, so a scan comes out the same quality it went in.
Nothing leaves your machine, which is unusual for this
Recognition is heavy, so almost every free service does it on a server, which means uploading the document. Think about what people scan: passports, bank statements, medical letters, contracts, certificates. It is close to the worst category of file to hand to a stranger.
The engine here downloads once, about four megabytes, from this site rather than anybody else's, and runs on your machine. You can watch the network tab while it works. Nothing is sent and nothing is stored.
The cost is time. A page takes a few seconds, so a forty page document is a couple of minutes rather than instant. It shows you each page as it finishes so you can stop early if the scan is not good enough to be worth continuing.
What it reads well, and what it does not
Printed text on a clean scan is read very accurately, and the confidence figure per page tells you what it made of yours.
A faint, crooked or low resolution scan is much worse, and it says so rather than handing you a confident-looking mess. Rescanning at 300dpi, flat and straight, does more for the result than any setting here.
Handwriting is not supported. Nothing free does it reliably and the paid services are poor at anything but neat block capitals.
Check your figures. Recognition confuses 0 with O, 1 with l and 5 with S on poor scans, and those are exactly the characters in the things people use this for: reference numbers, heat numbers, part numbers and amounts.
If your PDF already has text, do not use this
The tool checks and tells you. A PDF straight out of Word or a drawing package already contains perfect text, and recognising a picture of it can only be worse than what is already there.
The test is simple and you can do it yourself: try to select a line of it. If the cursor picks out words, there is text and you need nothing here.
Common questions
Why can I not copy text from my scanned PDF?
Because there is no text in it. A scan is a photograph of paper, so what looks like writing is a picture of writing. Recognition reads the shapes of the letters out of the picture and puts real text back, which is what this does.
How do I make a scanned PDF searchable for free?
Drop it in and press read. Every page is drawn, read, and the words are put back into the document invisibly. The file still looks exactly as it did and can now be searched, selected and copied from. No account and nothing uploaded.
Does the scan still look the same afterwards?
Yes. The pages are not touched. The text is drawn on top at zero opacity, so it is present for searching and selecting and paints nothing at all. Nothing is rasterised or re-encoded, so the file stays the quality it was.
How accurate is it?
On a clean 300dpi scan of printed text, very. It reports a confidence figure per page so you can see what it thought of yours. It confuses 0 with O and 1 with l on a poor scan, so check reference numbers and amounts.
Can it read handwriting?
No, and nothing free does it reliably. This reads printed text. Handwriting needs a different kind of model entirely and even the paid services are unreliable on anything but neat block capitals.