Because PDF stores where things are drawn on a page, not what they mean. There are no paragraphs or styles to recover, so a converter has to reconstruct that structure by inference. Simple text documents come back cleanly. Complex layouts with columns, tables and floating images are where it degrades, and there is no converter that fully solves this.

It depends on the tool and each page says which. Image conversions and simple text formats are processed in your browser and never leave your device. Video, audio and some PDF work needs real processing power and runs server side, where the file is used for that one job and deleted afterwards.

For images, resize before you compress, since halving the dimensions removes roughly three quarters of the data before any quality loss happens. Then use the Image Compressor and step the quality down gradually. For video, the Video Compressor works on bitrate, which is the setting that actually governs size.

Because it guesses. It strips leading zeros from anything that looks numeric, converts long numbers to scientific notation, and reads certain text as dates. The fix is to import rather than open, and to set the affected columns to text during the import step, which the guides walk through.

Only with OCR, which recognises characters in an image rather than reading stored text. Quality depends heavily on the scan: clean, straight, high resolution pages work well, while photographed pages at an angle do not. Always proofread the output, particularly numbers, where a mistaken digit is easy to miss.

For browser based tools the limit is your own device memory, so files in the tens of megabytes are usually fine and very large ones may slow the tab. Server side conversions have a stated limit shown on the tool page itself.