Tracked changes are the redline edits Word records when Review → Track Changes is on. They show up in word/document.xml as <w:ins> and <w:del> elements with attributes that record who made the edit and when. Most students don't use this feature deliberately — but when it shows up in a submission, it's a strong, file-structural signal about how the document came together.
This page covers what tracked changes mean for authenticity reading, what their absence means, and how the signal compares to other edit-history signals.
What tracked changes actually record
Each <w:ins> (insertion) and <w:del> (deletion) carries:
w:id— a sequence number within the document.w:author— the display name of the editor when that change was made.w:date— ISO-8601 timestamp.
So a tracked-changes document is effectively a serialized list of "Person X inserted phrase Y at time T". When you walk the list in time order, you reconstruct the actual writing process: which paragraphs came first, which were revised, what got cut, who introduced each phrase.
Word will also keep <w:moveFrom> / <w:moveTo> pairs for content that was relocated, and <w:rPrChange> / <w:pPrChange> for formatting changes. The detail level is high.
Why this is hard to fake
A student who wants to fabricate tracked changes faces a real engineering problem:
- They have to enable Track Changes from the start. Word won't retroactively backfill an edit history; it only records changes from the moment Track Changes is turned on.
- They have to actually do the writing in the document — typing, deleting, reordering, revising — over a period of time that matches the timestamps. Pasting a finished essay in once with Track Changes on produces a single
<w:ins>for the whole essay at a single timestamp. That's not what authentic edit history looks like. - The timestamps have to be plausible. Word's timestamps are local-time + offset; if a student tries to spread them out by changing the system clock, the offsets become inconsistent.
There's no command in Word to "generate plausible tracked changes for this finished essay." There are tools that can edit the underlying XML directly, but the result looks artificial: round-numbered timestamps, suspicious authorship strings, unrealistic insertion/deletion ratios.
In practice, fake tracked changes in student work are rare and usually obviously wrong. Real tracked changes are very strong positive evidence.
How to read a tracked-changes timeline
When you scan a docx with Autotend Forensics and the report shows tracked changes, the relevant questions are:
1. Time spread
Do the timestamps cluster in a realistic window? A 1500-word essay should produce edits across at least 30 minutes — typically several hours, often multiple sessions. All edits inside a single 60-second burst is unusual.
2. Author consistency
Is w:author always the same name? It usually should be. If the document was passed between collaborators legitimately, you'd see two or more names; otherwise, a single name is the expected pattern. A single-author paper showing two authors on <w:ins> is something to ask about.
3. Insertion-vs-deletion ratio
Real writing is messy. The ratio of <w:ins> to <w:del> typically reflects a process: drafting (lots of insertions, few deletions), then editing (more deletions, sentence rewrites). A document with zero deletions across hundreds of insertions usually means the student deliberately turned Track Changes on at the end and re-typed the finished paper into a new file — observable as a single insertion burst.
4. Insert order vs final order
The w:id order tells you the order edits were made; the position in the final XML tells you where they ended up. Real writing tends to be sequential — paragraph 1 mostly gets written before paragraph 5. A pattern of edits scattered across the document in non-sequential order (someone bouncing around to fill in pre-marked sections) is consistent with both legitimate revision and working from a pre-existing outline / structure, including an AI-generated outline.
What the absence of tracked changes means
It means nothing about authenticity. Most students don't use Track Changes. It's an opt-in feature, and most word processors don't have it on by default.
If a docx submission has no tracked changes, that's the default state. The absence is uninformative. Don't read it as a negative signal — the revisions.xml file and other edit-history signals are what matter when tracked changes aren't present.
The strong positive signal
When tracked changes ARE present and the timeline reads cleanly — multi-session, realistic edits, consistent author, realistic ins/del ratio — that's one of the strongest authenticity signals in document forensics. It's hard to fake, and it tells a very specific story about how the document came together.
Some instructors require Track Changes on for major assignments specifically to surface this signal. The student gets one minor friction point ("turn this feature on"); the instructor gets a structural record of the writing process that's an order of magnitude more useful than any prose-style detector. If this fits your class, the methodology page on edit-history signals walks through the full set of edit-history detectors that complement Track Changes.
What Autotend Forensics surfaces
The scoring engine reads tracked changes and produces several observations:
- Tracked changes present, multi-session distribution → authenticity positive (decreases AI-suspicion).
- Tracked changes present, single-burst → ambiguous; flagged for human review.
- No tracked changes, paste-shaped insertion patterns in revisions.xml → potentially indicative of copy-paste workflow; depends on other signals.
The scoring engine never asserts a positive verdict from tracked-changes data alone, and it never asserts a negative verdict from their absence. Both are weighted appropriately into the per-paper rolling baseline.
When this signal is most useful
- Major assignments where students can plausibly be asked to leave Track Changes on.
- Re-submission cases where the question is whether the student themselves revised the file.
- Comparison across drafts — earlier and later drafts of the same essay can be compared via their tracked-changes histories.
The signal is less useful for one-shot reflection essays, in-class writing, or any context where Track Changes wasn't an expectation. For those, fall back on the structural signals deep-dive.