Start with two comparable sources
Paste the original on the left and the revision on the right, or choose UTF-8 text files. Keep the same scope on both sides: a whole document against one paragraph creates a large but unhelpful deletion. PDF and Word files need text extraction elsewhere first. If you use OCR for a scan, verify names, numbers and line breaks before treating an OCR difference as a real edit.
Use lines for structure, characters for detail
Line mode shows which paragraphs or code lines changed. Character mode is useful for short passages, punctuation, identifiers and small spelling changes. For example, “Delivery: Monday” versus “Delivery: Tuesday” is one changed line but several character changes. The labels below the result always state whether the counts are lines or characters; they are not a count of semantic edits.
Understand the whitespace policy
The tool converts Windows CRLF and old-style CR line endings to LF before comparison. It does not otherwise trim or normalize your text. Two spaces differ from one, capitalization matters, and a missing newline at the very end can be a change. Character mode counts Unicode code points, so a displayed emoji may count as several characters. Visually similar Unicode sequences may still differ.
Read the highlighted result
The result follows source order. Underlined plus-marked text is added; struck-through minus-marked text is removed. Unmarked text is unchanged. These redundant markers help when color alone is hard to distinguish. Swapping sides reverses what counts as added or removed and clears the old report. Editing either input also clears the result until you compare again.
Keep the report with appropriate care
Download comparison JSON to preserve the full change list, counts and newline policy. It includes unchanged passages too, so it can reveal much more than just the edited sentence. Keep it with the two source drafts and share only with intended recipients. The report is not a patch file, an audit signature or a formatting-preserving Word document.
Handle a large difference honestly
Each side is limited to 100,000 JavaScript characters and 400 KB of UTF-8. Very different documents can hit a calculation or readable-output limit even below that size. Compare smaller sections or switch to lines; the tool does not silently omit the rest. Clear or leave the page to discard its in-memory workspace.