What you are comparing, and how it gets split up
For prose, the unit of comparison is the paragraph rather than the line — any block of text with a blank line either side of it. That single choice is what makes a document comparison readable: a paragraph you rewrapped to a different width stays one unit, so it is reported as one edited paragraph with the changed words marked inside it, rather than as a run of deleted lines followed by a run of added ones.
A reworded sentence is reported as one modified unit with the specific words that changed highlighted inside it, rather than as a deleted paragraph followed by an added one. So the output reads as a list of edits at roughly the granularity a person would describe them: this sentence changed, that paragraph moved, this section is new.
Comparing Word (.docx), OpenDocument (.odt), and RTF documents without office suites installed
Drag two .docx, .odt, or .rtf files straight onto the left and right panels to compare drafts without needing Microsoft Word, Office 365, or LibreOffice installed. The browser reads the document containers directly in memory, extracts the raw paragraphs, and displays the differences side by side immediately.
Plain text, Markdown, source code, and configuration files drop in the same way. Because the file is processed entirely within your browser using standard client-side APIs, dropping a document does not upload it to any cloud server, and your original file remains untouched on your disk. The two files do not need matching names or extensions.
What extraction discards is styling: fonts, colors, bold weights, tables, footnotes, tracked changes, and margin annotations are omitted from the comparison stream. What you keep is the raw text, which for "what words actually changed between draft A and draft B" is almost always the target. If visual styling or table cell formatting is the primary question, native word processor diffing is appropriate; for detecting textual revisions, paragraph-level text extraction is faster and clearer.
Comparing text from two PDFs online
Comparing two PDF documents side by side works by extracting the underlying text layer glyph by glyph in your browser:
- Vector and digital PDFs: Two exports of the same report, presentation, or publication compare cleanly. Words, paragraphs, and numbered items are identified from their bounding coordinates.
- Scanned PDFs vs text-layer PDFs: A scanned document is simply a series of raster images wrapped in a PDF container. Because there is no underlying text stream, drag-and-drop extraction cannot read it. For scanned documents, run optical character recognition (OCR) first, then copy and paste the resulting text.
- Reflowed text gotchas: A document that was reflowed across different page margins or layout widths can introduce line breaks in new positions. Wrapped lines are preserved rather than guessed at, ensuring the comparison tool never invents text that was not in the original document.
For formats that do not support automated browser extraction — such as legacy binary .doc, Google Docs, spreadsheets, or slide decks — simply select all (Ctrl+A or Cmd+A), copy, and paste the text directly into the respective panel.
Reordered sections and moved blocks
Moving a section is where a document comparison usually goes wrong, and it is worth knowing which answer you are getting. In Smart mode a relocated paragraph is recognised as relocated: one change, with a connector linking where it was to where it now sits, costing almost nothing against the similarity score. That is normally what you want when reviewing a restructured draft — the content survived, and saying so is more useful than reporting it as a deletion and an unrelated addition.
Strict mode reads both documents top to bottom instead, so a moved paragraph appears as removed from one place and added in another, and the score drops. Use it when sequence is the thing you are checking: a procedure whose steps must stay in order, a numbered set of terms, an argument that depends on what came before it.
Reading both numbers together is quick and tells you more than either alone. A high Smart score with a low Strict one means the document was reorganised rather than rewritten. Both low means substantial new writing. See moved and reordered blocks for details on the matching algorithm.
Quieting the differences you do not care about
Two things routinely fill a prose comparison with noise. The first is capitalisation — a sentence that got recapitalised after an edit, a heading that was title-cased. Turn the Case sensitive switch off and those stop registering, without your text being touched.
The second is text pasted from a word processor or a web page, which often carries characters that look ordinary and are not: non-breaking spaces, curly quotes where the other side has straight ones, en dashes where the other has hyphens. These are genuine textual differences and the comparison is right to flag them, but they are rarely what you were looking for. If one side came from a different source than the other, expect some of this.
The To sentence case transform is the more targeted option for drafts: it normalises capitalisation rather than flattening it, so sentence-initial capitals survive while a recapitalised line stops reading as a change. It edits both panels and there is an undo beside it. Whitespace and case goes through all of this in more detail.
Confidential documents and legal privacy
The comparison runs in a Web Worker inside your own browser. There is no server here that receives your text and no endpoint that could — which is what makes it reasonable to paste a contract, a patient record, an unpublished manuscript, or an internal memo into it. That is a property of how the tool is built rather than a promise about who is trusted with the contents. Read our guide to client-side private comparison to learn how to verify this in your browser's Network tab.
Two consequences worth knowing:
- Saved comparisons: A comparison you explicitly save is kept in your own browser storage on that device, not on a server, so it is readable by anyone with access to that browser profile.
- Share links: A share link encodes the comparison in the part of the URL after the #, which is never sent to a web server by a browser — but it is still a link, so anyone you send it to can read the text, and it may sit in your history or a chat log.
What a document comparison cannot settle
It tells you what differs. It cannot tell you whether the change was an improvement, whether a clause still means what it did, or whether a rewritten paragraph says the same thing in different words — a heavily reworded sentence and a sentence whose meaning was reversed can produce a similar-looking row.
For anything with consequences — a contract, a filing, a medical or legal record, a published statement — use the comparison to find the changed passages and then read those passages yourself. Locating the differences is the part it does well, and the part that takes the longest by hand.