Are Tracked Changes and Comments in Word Discoverable?

The page that prints and the revisions beneath it are both part of the file a court can order produced.

Tracked changes and comments in a Word file are metadata that federal courts have treated as subject to preservation and disclosure, and they stay in the file until accepted, rejected or deleted.

A 2007 Florida Bar Journal article called the track changes tool "the most vulnerable part of the Microsoft suite of applications," because, once it is turned on, Word saves every modification in a form that could let the opposing side discern "exactly what was in the mind of the document drafters throughout negotiations and pleadings." Most writing on the subject has since treated that exposure as a cleanup chore. I treat it as evidence.

The record is precise. Word notes who made each change and when, and a comment, placed in the margin beside the clause it concerns, can record why. The first question for counsel is therefore a practical one: in what form did the client's people actually exchange the drafts that matter, since each form carries a different portion of the record? The research section below walks through where that record survives and where it is lost.

What does a Word file carry beneath its visible page?

A native Word file carries its tracked changes, comments, author names and revision history inside the file itself, so whatever is sent or collected brings that layer with it.

Before a collected Word file is treated as a finished page, three questions settle what it actually contains:

1. Is the item a native .docx file, or only a PDF or image rendered from it?
2. Does it still hold tracked changes or comments that were never accepted, rejected or deleted?
3. Do its file properties record authors, a revision count or total editing time?

In the Florida Bar Journal of October 2008, David Hricik and Chase Edward Scott examined their own unfinished article and found, in File/Properties, that it stood at its 44th revision with 205 minutes of total editing time, a record composed by the software without any request from the authors. They observed that track changes "creates a record of every change made to a document," that Word may not display that record on screen unless a particular option is enabled, and that comments are embedded in the file in such a way that they accompany it whenever it is e-mailed. The record is kept. It is merely not shown.

A fuller inventory had been set out in 2007 by Nicole O'Neal, also writing in the Florida Bar Journal, who listed among the things metadata can routinely reveal the original authors, the firm name, the computer name and file location of the drafting machine, revisions and versions, hidden text, drafter initials, undo/redo history and, placed at the very end of the catalogue as though of no particular consequence, "comments on the document, and tracked changes." She also noted that federal courts have found a duty to preserve metadata under Rule 34 and a duty to disclose it under Rule 26, with sanctions where parties failed to comply. Readers practicing in that state can set her forecast beside what Florida's 2026 e-discovery rules change.

The common assumption is that a document shows what it contains. It does not. Office saves changes, annotations, comments, previous versions and even previously removed text automatically, and switching Track Changes off leaves every existing revision exactly where it was. Metadata, in the e-discovery sense, is information about a document that is not part of its visible content, and for a Word file that category includes the revision history itself.

An analysis of 5 sources shows a consistent arrangement: the markup layer is formed by the software, carried inside the file, and delivered with it to whoever opens, sends or collects it. For a reviewer, the consequence is practical. A page read only as rendered is an incomplete document.

Reading the rest of it need not be priced like manual review. Relevant e-Discovery's AI-assisted review runs at cents per document, against the dollars per document and some $19K/GB of manual review, a difference a firm can measure on its own matter. In my view, that is what moves the hidden layer from an afterthought into the ordinary review queue, where the drafter's earlier sentence, struck through and never displayed, waits beside the clean one that was sent.

Can stripped or overwritten revision history be recovered once litigation starts?

Rarely from the stripped copy itself, but sometimes from a cloud original or a version preserved under hold, and any edit so recovered should trace to the exact exhibit it came from.

The question is composed of two parts, of which the first concerns the copy in hand and the second concerns everything that stood behind it. On the copy, the answer given in a 2022 Microsoft Word user forum, to a recipient who could find no revision history in a document sent to them, was flat: "you can only see what is there. If it is gone, then it is gone." The two causes that recipient had identified, a paste of the whole text into a fresh document or a pass through Document Inspector, each leave a file in which the earlier states no longer reside.

A second reply in the same thread placed the matter elsewhere. Where the master copy lives in SharePoint or OneDrive and the requester holds permission, File, Info, Version History will display the prior states of the original, so that a scrubbed copy does not prove the history was destroyed at the source. The stripped file is one object. The cloud original is another.

That original is itself unstable. Many "attachments" in Microsoft 365 are links to SharePoint or OneDrive files that continue to change after the message is sent, and a collection made months later returns the current state rather than the state that was shared. Older versions are overwritten once a library's version setting is exceeded, usually 500 out of the box; a litigation hold stops the overwriting, but only from the moment the hold is placed, and audit log retention defaults are often shorter than litigation timelines. Exporting every version is possible and still does not establish which one was sent.

Fragility runs in the other direction as well. Allison Shields, on an American Bar Association redlining panel, described collaborating on documents where all changes were supposed to have been accepted and tracking stopped, yet "some remnants remained." Removal is not reliable. Neither is survival.

For a small firm, recovery therefore depends on timing and location far more than on technique, and the places a collection can silently fall short, several of which are set out in six things one-click cloud collection tools miss, should be weighed before the hold letter goes out rather than after. When a prior version or a surviving remnant does appear, our review outputs link back to the exact exhibit they came from, so an attorney can confirm with one click that the fragment is genuine before it is quoted in a motion, which leaves the harder matter of reviewing such fragments at volume on a single, messy collection.

How should a small firm review tracked changes and comments defensibly?

Preserve the native files as immutable, hashed originals with an append-only audit trail, review the markup layer as text in its own right, and gate privilege before anything is produced.

The workflow I recommend is arranged in three steps, each placed so that the one after it can rely on what it has already fixed in place:

1. Preserve the native files exactly as collected, with a content hash on each original and a documented chain of custody, before anyone opens them for review.
2. Extract tracked changes, comments and any captured prior versions so that each can be read, searched and coded as text, rather than inferred from how the page renders.
3. Screen the comment threads and revisions for privilege, and let nothing pass to production until that screen has been applied.

The first step carries the weight of the others. A struck-through clause recovered from a draft is worth only as much as the proof that the draft was not altered after collection, and that proof is composed of immutable originals, content hashing, an append-only audit trail and documented chain of custody, the arrangement our process rests on so that it holds up if opposing counsel challenges it. Without that spine, the most revealing comment in the collection becomes a dispute about handling.

The second step is where the hidden layer is either read or lost. A reviewer shown only the rendered page cannot code what the page does not show. The layer is also changing in composition: AI drafting add-ins now write their edits into Word as native tracked changes and post their explanations as replies inside comment threads, so that a revision history may increasingly hold entries made by software at a person's request, placed beside the human ones and requiring the same attention.

The third step matters because comments are where reasons get written down. The privilege screen should be fail-closed, meaning a document that has not cleared it cannot be produced by default, a design discussed at length in designing a privilege gate so privileged ESI never reaches review. Where those comments may be privileged, I would also insist that processing run single-tenant, or inside the firm's own AWS account under its own keys, with no vendor retention and no model training, so that nothing fed into review trains anyone's model or waives privilege.

None of this requires an enterprise deployment. Relevant e-Discovery is built to be operable and affordable for a solo or small firm on a single messy matter, the segment that enterprise platforms such as Relativity and Everlaw price out and cannot serve without an administrator. The fairest test is concrete. Bring a messy collection and a hard question, perhaps the one about which draft of a clause a counterparty actually saw, and watch the system read, code and cite that firm's own evidence rather than someone else's case study.

What should litigators expect from tracked changes in the next two years?

Expect the markup layer to move from a cleanup chore to a contested category of evidence, with preservation timing and privilege review deciding most disputes over Word drafts.

Courts are already revisiting privilege. On a February 2026 HaystackID webcast, moderator Philip Favro called the FirstEnergy ruling "the most noteworthy decision on this issue since the Kellogg, Brown, and Root cases," and panelist Ruth Hauswirth held that relevance and proportionality govern information regardless of its type. A comment explaining why a clause changed sits squarely inside both arguments. It may be the most relevant line in a draft. It may also be the most privileged.

My own view is plain. Firms that collect natively, place holds before versions overwrite, and read comments as text will argue about what a draft meant. Firms that do not will argue about why the draft is missing, and that is the weaker position to defend before a magistrate judge who has asked what was preserved and when.