Short answer: Sometimes file metadata can hint that AI software touched a document, but it can’t settle the question. Metadata records which program created or last saved a file, plus dates the computer writes for itself. In October 2026 the Quebec publisher Boréal ended its relationship with novelist Thélyson Orélien after an outside review found “traces of AI markers in the metadata of certain documents”. He denies using AI to write the novel. Ordinary tools such as sync apps, converters and add-ons can change, strip or add metadata, so it is a weak witness either way. A steadier record is a dated trail of drafts that you timestamp outside your own computer as you go. A timestamp shows when a file existed, not who wrote it.
Disclosure: I’m Inna Hall, co-founder of TimeBinder, the file timestamping tool this guide mentions. This page is general information, not legal advice.
Table of Contents
What a Word or Scrivener file actually stores
Metadata is information about a file, kept alongside its text. A Word document (.docx) usually carries an author name, a “last modified by” name, created and modified dates, a revision count, total editing time and the name of the program that saved it. A PDF records similar details, including the program that produced it.
A Scrivener project is a folder of many small files, so most of its dates come from your computer’s file system and from Scrivener’s own records. When you compile or export, the new Word or PDF file gets fresh metadata of its own.
You can see what your own files record with the file date and metadata inspector, which reads the dates in your browser without uploading the file.
Why “Created” and “Modified” dates move on their own
These dates are written by whatever software last handled the file, using the clock on that computer. They can change when you:
- copy a file to a new drive or download it from email or the cloud,
- let a sync app (Dropbox, OneDrive, iCloud) restore or re-save it,
- convert it, for example from .scriv to .docx or .docx to PDF,
- open it in a different program, or run an add-on or grammar tool over it.
Anyone can also edit these fields on purpose with free tools, or by changing the computer’s clock. So a date inside a file tells you what the software wrote, not necessarily when the words were written.
Why this matters now: the Boréal case
On 21 October 2026 Les Éditions du Boréal said it was ending its relationship with Thélyson Orélien and would not reprint his novel C’était ça ou mourir, which had been withdrawn from the Prix Goncourt list the month before. The publisher said it had gathered all the working versions of the novel and commissioned an outside analysis, which found traces of AI markers in the metadata of some documents. Boréal also said it still did not know to what extent AI had been used, and might never have a definitive answer. Orélien has repeatedly denied using AI to write the book (The Guardian, 22 October 2026; Boréal statement, in French).
We don’t know what the files contained, and this article doesn’t take a side. What the case shows is that, once an accusation is made, publishers turn to the files themselves, and the details inside them can decide the outcome. Earlier in the dispute, the publisher and author had tried to put together a record of the writing process and it did not settle the question.
For authors, the lesson is practical. A trail of drafts made and dated as you went, with dates you don’t control, is much easier to explain than metadata you have to interpret after the fact. See also what to do about a manuscript facing an AI accusation.
File metadata and AI: what it can and can’t tell a publisher
File metadata and AI use are only loosely linked. Metadata can show:
- which program created or last saved a file (for example, a file saved straight out of an AI tool may say so),
- dates and editing time, as recorded by the computer that saved it,
- sometimes, traces left by add-ons or other tools.
It can’t show:
- who typed the words, or how much of the text came from a tool,
- whether a marker came from writing, from research, or from a tool that only reformatted or checked the file,
- whether the details were changed after the fact.
Metadata can raise a question. On its own, it can rarely answer one, in either direction. AI detectors have similar limits; see why working drafts and timestamps matter more than AI detectors.
Metadata vs version history vs an independent timestamp
| File metadata | Version history (Word, Google Docs, Scrivener snapshots) | Independent timestamp | |
|---|---|---|---|
| Who controls the date? | Your computer and software | You, or your cloud account | A public record outside your control (the Bitcoin blockchain) |
| Can it be edited later? | Yes, easily | Sometimes, by the account owner or provider | No. The record stays fixed, and any change to the file stops it matching the recorded fingerprint |
| What it shows | What the software wrote about the file | How the text changed over time | That this exact file existed no later than a recorded time |
| What it doesn’t show | Who wrote it, or when the words were written | That the dates weren’t changed | Who wrote it, or whether AI was used |
The three work best together. Version history shows how the book grew; an independent timestamp pins each milestone to a date nobody can quietly move.
How to keep a draft trail that holds up
- Save milestone copies. At each stage (outline, first draft, each revision pass, the version you submit), save a copy with the stage and date in the file name. For Scrivener, a zipped backup or a compiled export is a single file that is easy to keep.
- Timestamp each milestone. Record a fingerprint (a hash) of each copy with an outside timestamp. With TimeBinder the file is hashed on your own computer and only the fingerprint is recorded; your manuscript is never uploaded. A Pooled TimeBind covers up to 10 files and is usually recorded within 24 to 48 hours, and often sooner, depending on Bitcoin network activity.
- Keep originals unchanged. Store the timestamped copies and their certificates together, and keep working on a fresh copy. Opening and re-saving a timestamped file changes it, so it would no longer match its fingerprint.
- Keep the rest of the trail too. Notes, research files, emails to your editor and printed drafts all help show how the work developed.
For step-by-step instructions, see how to show a Word document or Scrivener project existed before a date and using AI as a writing tool and showing which words are yours.
Keep a dated trail of your drafts. Timestamp each draft, note and manuscript as you go, so you can show when every version existed. See how draft timestamping works →
Files are hashed locally on your computer and are never uploaded.
The honest limit: a timestamp shows that a file existed, unchanged, no later than a recorded time. It doesn’t show who wrote it, how it was written, or whether AI was used. It is one part of a record of your process, not a verdict.
Frequently Asked Questions
Can metadata prove I didn’t use AI?
No. Metadata only records what software wrote about a file, and it can be changed. Clean metadata doesn’t show that no AI was used, and an odd tag doesn’t show that it was.
Can someone edit a document’s created date?
Yes. Created and modified dates can be changed with free tools or by changing the computer’s clock, and they also change when files are copied, synced or converted.
Does a grammar tool leave traces in metadata?
It can. Some add-ons and editing tools leave their own details in a file. The safest step is to check what your own files record before you send them.
What does a timestamp show?
That a specific file existed, unchanged, no later than a recorded time. It doesn’t show who made the file or how it was written.
Does TimeBinder read my manuscript?
No. Each file is hashed on your own computer and only the fingerprint is recorded. Your manuscript is never uploaded or stored.
Plan details were checked against timebinder.io/pricing on 30 October 2026. News details were checked against the sources linked above on 30 October 2026.
