You can’t prove that AI was never used on a piece of writing, and nobody else can prove that it was, including a detector. What you can show is how your work grew. Gather what already exists: version history in Google Docs or Word, outlines, notes, research history, and emails with teachers, editors or beta readers. Export the key versions as PDFs and write a short, dated timeline of how you drafted the piece. Say plainly which tools you used, from spell-check to AI. If you’re a published author, post one calm statement and send the evidence privately to your publisher, agent or retailer. For future work, timestamp each draft as you go: a TimeBind shows that an exact draft existed no later than a recorded time. It doesn’t show who wrote it, but a steady trail of dated drafts is hard to fake after the fact.
Disclosure: I’m Inna Hall, co-founder of TimeBinder, the file timestamping tool mentioned in this guide. This page is general information, not legal advice. If you are a student, your institution’s own policy and process come first; if you are a published author, so do your contract and your publisher’s process.
Start a dated draft trail with your next draft. Timestamp each version as you finish it. Each TimeBind shows that exact file existed no later than a recorded time; it doesn’t show who wrote it or whether AI was used. The Free plan gives you 1 Token a month (1 Pooled TimeBind, up to 10 files).
Table of Contents
- Why AI detectors get it wrong
- Gather what already exists
- How to prove you wrote something and not AI in Word or Google Docs
- How to explain your process
- If you’re an author accused of using AI
- For next time: timestamp drafts as you go
- What each kind of evidence shows
- Where to get help
- Which plan fits
- What a timestamp does and doesn’t do
- FAQ
Why AI detectors get it wrong
AI detectors give a probability, not a finding. They look for patterns such as predictable word choice, and plenty of human writing has those patterns: formal essays, writing by people working in a second language, and short answers that follow a set structure. Different detectors often give different scores for the same text. That is why a detector score is weak evidence either way, and why a “human” score from a second tool won’t settle things. The same goes for screenshots of detector scores posted in reviews or comments about a published book: they are a guess about text, not evidence of how it was written. We go into this in more detail in why working drafts matter more than AI detectors.
Gather what already exists
Before you write a single reply, collect the records you already have. Don’t edit or tidy anything: you want the history as it is.
- Version history. Google Docs keeps a detailed history of edits. Word keeps versions for files saved to OneDrive or SharePoint, and AutoRecover files may exist on your computer.
- Outlines and notes. Paper notes, phone notes, a mind map, a plan you emailed yourself.
- Research history. Browser history, library loans, saved PDFs and highlighted sources.
- Messages. Emails or chats where you talked about the topic, asked for an extension or shared a draft, including notes from editors, agents and beta readers. These are held by other people, which helps.
- Earlier work. Past assignments or articles in your own voice can help show your usual style.
How to prove you wrote something and not AI in Word or Google Docs
Version history is usually the strongest record you have, so present it clearly:
- Google Docs: open File > Version history > See version history. Name the key versions (first outline, first full draft, after feedback, final) so they are easy to find.
- For each named version, use File > Make a copy or download it as a PDF, so you have fixed copies outside the live document.
- Word: if the file is on OneDrive or SharePoint, open File > Info > Version History and save copies of the main versions. If it was only on your computer, check for AutoRecover files and any copies you emailed or uploaded.
- Put the exports in one folder with a short timeline: date, what changed, roughly how long you spent.
- If you write in Word or Scrivener, see proving a Word or Scrivener file existed before a date for what those files record about time.
Remember that version history lives inside your own account. It is good evidence of a process, but others can’t always inspect it directly, which is why it helps to pair it with other records. Our page on what cloud version history can and can’t show covers the limits.
How to explain your process
If you are asked to meet a teacher, editor or client, stay calm and be specific. Walk them through the timeline: where the idea came from, what you read, how the structure changed, and which parts were hardest. Offer to talk through any paragraph in your own words. Be upfront about tools: if you used spell-check, Grammarly or a translation tool, say exactly what it changed. If you used AI for brainstorming or research questions, say that too; our guide to using AI as a writing tool and showing which words are yours explains how to separate the two.
If you’re an author accused of using AI
For published authors, the accusation usually comes after release: in reviews, on social media, or from a retailer or publisher. The evidence is the same as above (version history, early drafts, notes and emails with your editor), but the audience is different. Post one short, calm public statement, then send your evidence privately to your publisher, agent or retailer.
Our step-by-step guide for authors accused of using AI covers what to say, what goes in the evidence pack, what to do if the question is about your cover, and what to skip. If you used AI tools along the way, our guide to using AI as a writing tool and showing which words are yours covers how to be open about it. For your next book, see how to show a manuscript existed before an AI accusation.
For next time: timestamp drafts as you go
Once you’ve been through this, it’s worth building a habit. Your file is hashed on your own device, which means TimeBinder works out a unique fingerprint of the file, and only that fingerprint is anchored to the Bitcoin blockchain. Each TimeBind shows a draft existed no later than its time. It doesn’t show who wrote it or whether AI helped, but a series of dated drafts shows the work developing.
- Keep separate files for each milestone: outline, first draft, each major revision, the version you submit.
- TimeBind each milestone as you finish it, together with that stage’s notes. One TimeBind covers up to 10 files.
- Never re-save a sealed file. Work on a copy; even an unchanged re-save can change the fingerprint.
- Keep each Proof of Time Certificate next to the files it covers.
Between milestones, you can also keep a daily draft log of word counts and file fingerprints; its dates come from your device’s clock, so TimeBind the milestones for a recorded time.
Start timestamping each draft as you write →
What each kind of evidence shows
| Evidence | Shows the work developed over time | Who holds the record | Who it tends to convince | Weak spots |
|---|---|---|---|---|
| Google Docs version history | Yes, in detail | Google, inside your account | Teachers, editors, publishers | Others can’t easily inspect it; it’s tied to your account |
| Word version history / AutoRecover | Sometimes | Microsoft / your device | Teachers, editors | Patchy if autosave was off |
| Notes, outlines, research history | Partly | You | Anyone, as context | Easy to dismiss as recreated |
| Emails with editors, agents or beta readers | Partly, at the points you shared drafts | The other people and their email providers | Publishers, agents, retailers | Only covers what you sent, and when |
| Designer process files (book covers) | Yes, for the cover | Your designer | Readers, retailers, publishers | Depends on the designer keeping them |
| AI detector “human” score | No | The detector vendor | Rarely anyone for long | Detectors disagree; false positives happen |
| Timestamped drafts (TimeBind) | Yes, if you timestamp along the way | Bitcoin record. ASAP TimeBinds can be checked independently from the certificate and the blockchain; Pooled TimeBinds through TimeBinder | Anyone who checks the file against its certificate | Only helps if done before the dispute. Doesn’t show who wrote it or whether AI was used |
No single row settles the question. Together, records made at the time tell a consistent story.
Where to get help
Students can usually get free, confidential help from a student advocacy or student union service, and should read their institution’s academic integrity policy before any meeting. Professional and published writers can often get advice through a writers’ organisation or society of authors in their country, and should tell their agent or publisher early. If a contract, payment or your reputation is at stake, get advice from a lawyer where you live.
Which plan fits
The Free plan is $0 and gives you 1 Token a month. A Pooled TimeBind costs 1 Token, covers up to 10 files and is usually recorded within 24 to 48 hours, and often sooner, depending on Bitcoin network activity. An ASAP TimeBind costs 3 Tokens and gets its own Bitcoin transaction straight away. Checking a file is always free, and no credit card is needed for the Free plan.
If you TimeBind more often, Essential is US$20 a month for 5 Tokens ($4 per Token, against $5 pay-as-you-go) and Expert is US$50 a month for 15 Tokens (about $3.33 per Token). Paid plans replace the Free plan’s monthly Token, so Essential adds 4 Tokens for US$20 and Expert adds 14 for US$50. See pricing.
What a timestamp does and doesn’t do
TimeBinder anchors each file’s SHA-256 fingerprint to the Bitcoin blockchain and gives you a Proof of Time Certificate. That is a dated record of the file: it shows the file existed no later than the anchor time, and any later change to the file is detectable. Your file is hashed in your browser and never uploaded.
It doesn’t show who created the file, whether its contents are true, or whether AI was used to make it, and it can’t tell you how a court or anyone else will treat it. It also doesn’t cover versions you didn’t TimeBind. Pooled TimeBinds are verified through TimeBinder; ASAP TimeBinds can also be verified independently from the certificate and the Bitcoin blockchain. If you are a published author, see also proving a manuscript existed before an AI accusation and timestamping for writers.
Frequently Asked Questions
Can I prove I didn’t use AI?
Not absolutely, and nobody else can prove that you did use it either. What you can do is show a believable, dated trail of your own drafting: versions, notes, sources and a clear account of any tools you used.
Is Google Docs version history enough?
It’s often the most useful thing you have, because it shows the text growing over time. But it sits inside your own account and can be questioned, so pair it with exported versions, notes and messages held by other people.
Can a timestamp prove my writing isn’t AI?
No. A timestamp shows that an exact file existed no later than a recorded time. It says nothing about how the file was made or who wrote it. What helps is a series of timestamped drafts that shows the work developing.
How should an author respond to readers accusing them of using AI?
With one short, calm public statement about how you wrote the book, then send your evidence privately to your publisher, agent or retailer rather than arguing in comments.
What should I bring to an academic integrity meeting, or send my publisher?
A one-page timeline of how you wrote the piece, exports of key versions, your notes and outline, your sources, emails with anyone who saw drafts, and a short list of any tools you used. Students should read their institution’s policy first and ask whether they can bring a support person.
Can I timestamp drafts I wrote months ago?
Yes, but the timestamp will show today’s date, not the date you wrote them. It still freezes those files as they are now, so later changes are detectable. From here on, timestamp each new draft as you finish it.
Plan details were checked against timebinder.io/pricing on 30 October 2026.
