Finding what actually changed between two versions
Comparing documents by eye is unreliable in a specific and dangerous way: you catch the large changes and miss the small ones. A reworded paragraph is obvious. A single altered number, a removed "not", or a changed date is exactly the sort of edit that matters most and is easiest to read straight past.
This is why comparison matters most on contracts and agreements. A document returned by another party as "the same with a couple of tweaks" deserves a check that is systematic rather than impressionistic, and the changes that get described that way are not always the ones that were made.
The same applies to your own versions. Establishing which of two drafts is the later one, or what a colleague changed while you were away, is faster to answer by comparison than by reading both documents and trying to hold them in mind.
What comparison can and cannot see
Text comparison finds differences in the words. That covers the overwhelming majority of meaningful changes and is what you want in almost every case.
It does not, by itself, catch changes that are not textual: a modified figure in an image, an altered chart, a changed signature, different formatting that changes emphasis. A document where a number lives inside a picture rather than in the text will compare as unchanged.
It also cannot compare a scanned document with a digital one in any meaningful way unless the scan has been through OCR, since one has text and the other has pixels. Comparing an original against a signed, scanned copy therefore needs OCR first, and the recognition errors will show up as differences that are not really differences.
Reading the results sensibly
Expect noise. Reflowed paragraphs, different line breaks, changed page breaks and reformatting all register as differences even when no word changed, and a document that has been through a different editor can show hundreds of them.
Work from most to least significant rather than top to bottom. Numbers, dates, names, defined terms and anything in a clause about liability, termination or payment deserve attention first; whitespace and punctuation almost never repay it.
And treat the comparison as a tool for directing attention rather than as a verdict. It tells you where to look. Whether a change matters is a judgement the software is not making for you, and on anything consequential the flagged sections still need reading properly.
Building comparison into how you handle contracts
The habit worth forming is comparing every returned document, not just the ones you are suspicious about. Comparison is cheap and takes a minute; discovering an altered clause after signature is neither.
Keep the version you sent, clearly named, so there is always a known baseline. The most common reason a comparison cannot be done properly is that nobody kept a definitive copy of what went out, and reconstructing it from an email thread is unreliable.
Where a document has been through several rounds, compare against the immediately preceding version rather than against the original. Comparing round four with round one produces a wall of differences that tells you nothing about what changed most recently, which is usually the question you actually have.
Agree how changes will be marked before a document starts circulating. A counterparty who returns a document with tracked changes, or with a covering note listing amendments, saves everyone the comparison entirely. Where that is not the practice, comparing is the fallback — but asking for marked changes at the outset costs nothing and is far more reliable than detecting them afterwards.