What is AuthorProof?

AuthorProof is a simply, easy to use app that I designed for authors to prove that they did not use AI to write. You can read my very long-winded Substack article about this—which I am about to freely cage off of in the more streamlined FAQs below.

How can I get it?

You can fill out this form here. I am in the process of enrolling in Apple’s developer program and fixing some final things. I hope to start sending the app out in about two weeks (late August.)

Who can use AuthorProof?

AuthorProof can be used on Mac or PC with Microsoft Word, as long as your version of Word works with .docx files. I am not currently working on a version of Scrivener or any other writing program, but please scroll down to Alternatives to AuthorProof for more information.

What does it cost?

It is free. I initially wanted to charge something nominal, but decided to give it away for free after 3 minorities in a row were accused of using AI. I don’t know if they are guilty or innocent: I do know that it is not the case that 100% of people using AI to write are minorities. Please note that if the app runs into problems, I will do my best to fix it, but ultimately that is on volunteer time.

How does it work?

A longer and more chatty explanation can be found in the Substack, but in brief: You open your Word doc. You open AuthorProof. You click Start in program. You write. You click “Done for now.” That’s it. The program works by keeping (on your computer) an archive containing a few different items. One file—which essentially logs how you have changed your Word document from the last time you worked on it—gets hashed with SHA-256 (a cryptographic algorithm), and 1) gets sent to a third-party time stamper and 2) it gets bundled with a bunch of other hashes and a fingerprint of this bundle gets permanently recorded to the Bitcoin blockchain. This means your actual content does not leave the privacy of your computer, but that in your archive, you will have record of a chain of certificates verifying your work at particular points in time. This file—a json file—can provide evidence of your work, but the program also provides a pdf “sister” document that is more palatable to read that effectively shows your timestamped work over time. You can click here to view a sister document of the Substack article as I originally wrote it. Note that the PDF shows when text was added, when it was deleted, and when it was altered in some way—and each session is timestamped. This is different than simple tracked changes version of your document because track changes wouldn’t denote exactly when edits were made. My method might also show evidence of, for example, you agonizing and repeatedly editing a section in your first chapter for months and then you deleting the entire section a year later (an AI wouldn’t do that). Your document archive will contain this json file, the sister document, and all your time-stamping certificates.

Can’t someone use AI to write while still using your program and fool us all?

I am aware that there are AI typing programs that type at a human rate and even make typos. If you think this is how humans right fiction, this is hilarious. Writing a novel does not consist of typing and fixing typos. It consists of writers hemming and hawing on the page, changing their minds, writing characters in or out, changing who the killer is, making massive shifts for editorial and substantial reasons. Could someone use AuthorProof to try to fake a record of work, say for example, them typing out content that ChatGPT gave them? This would require the person who wants to use AI but deny they are using AI to carefully spread out their work over a long period of time, working only in sessions that seem human-sized. I spent over 118 hours writing my last novel, spread out over 2+ years. I also made substantive editorial things that an AI wouldn’t do, like making a friendship a more important component of the story, or “increasing the sexiness” between two characters at the request of my editors. You could use AuthorProof to log your AI-written novel, but you would have to spend an insane number of hours logging hundreds of sessions, deliberately seeding human errors, and doing things to make it all seem real, like not having enough sexiness in the draft for 1 year, but then adding edits to increase sexiness—when the reality is, planning out such a performance and creating such a document would actually take more time than writing an actual novel. I do not believe that the sort of people who want to write with AI and not disclose this to readers will bother to do this. Bona Books published an excellent paper about how they did AI detection, and what they found is that people who want to pass off AI writing as their own are not criminal masterminds who are willing to take years to create elaborate paper trails.

How do these various things prove that I wrote something and not AI?

The sister document combined with the json file (the sister is just easier to read) is the strongest argument to show evidence of human behavior over time—it will take someone reading it carefully. Ah, but can’t someone fake that document? This is where the chain of certificates come in. To use a surprisingly apt analogy, let’s discuss the diary in Gone Girl. In the book, we read Amy’s diary, which shows the history of her marriage’s decline to the very last entry where she fears her husband Nick may kill her. Then we find out that Amy faked these diary entries, writing them all more or less at the same time, using different pens, to show a “history” of abuse and threat from her husband in order to frame him. But if she had to certify the way I’ve outlined above, this wouldn’t have worked. On day 1, I write my diary, and a notary stamps it, indicating the date and time (ie, third-party time stamping). On day 2, I write my diary entry, and the notary stamps, certifying the date and time of everything from day 1 and 2. So the stamp certifies not just today’s content, but everything that came before it. So you have a chain of trust. The user ultimately has all the certificates and could provide the json file and sister document to whoever is asking. Anyone with a computer can run SHA-256 themselves on the json file and make sure that this result matches the most recent hash in the hash log provided in the archive folder. If it matches, we have independent confirmation that that exact hash existed at the time of timestamping. If it doesn’t, this means someone altered the json file.

What does this have to do with Bitcoin?

Nothing, actually. The Bitcoin blockchain just happens to be extraordinarily useful. This is what occurs with the program in plain English: your hash—the hash of your “what did I accomplish today” file— is sent to a public ledger that is permanent and publicly verifiable. The blockchain exists simultaneously on tens of thousands of independent computers around the world: it is not the case that a single person can decide to delete it. The reason I did third party verification is that people can do things like manually change the metadata in their Word document to change “date created” and the like. If Bitcoin makes you itchy, just know that the process I just described doesn’t really have anything to do with Bitcoin as a currency. The blockchain is really useful because it is permanent and tamper proof. All this happens while maintaining your privacy—no one can tell what your hash says.

How did you write this app?

I used Claude Cowork. Please be aware that I’m a reasonable person who understands exactly how authors feel about AI—because I am one of those authors. My entire body of work was stolen by AI companies—all my books, even the German translation of my first book, all of my online writing no doubt as well. And I received no compensation for this. I am not part of the Anthropic settlement because of exactly when my first book was stolen. I have no idea and I am skeptical that I will ever see a settlement from OpenAI or Meta even though those companies knew they were stealing. I had ideas about this app years ago, but waited for years because I assumed someone, like a major company, would build this on their own. I put the idea out there publicly hoping someone else would run with it—someone who already knew how to program. I even contacted Microsoft (because honestly I think they are best placed to do something about this) but I received no response. I know many writers take an absolutist position on AI, which I totally understand. I don’t use AI to write text, or outline, or brainstorm or edit, or whatever because I believe I am better at doing these things than it is, and these are things that I enjoy doing that make up the backbone of writing. I would much prefer if Microsoft came out with something that did this, or if the Big 5 developed some kind of process, but right now it feels like there are no adults in the room and I keep hearing authors stress out about this and ask what they can do. So, I don’t know what to tell you. I saw a toilet overflowing, and I used some shit to prevent even more shit from coming out of said toilet. It is ironic; you needn’t point out the irony to me.

What happens if I am working and don’t have internet access?

The program will still function—you just won’t be able to get your timestamp certificates until the next time you click “done for now” on your project. So if you work offline on Monday but don’t have internet access till Saturday, you work will be logged in your json as occurring on Monday, but the timestamp itself will say Saturday.

Can I rename my Word document?

Yes, you can rename within the program itself, which is the preferred method. If you renamed your doc outside the program on your own, the program will be unable to find the document, but your archive will be safe. The program will just ask if the file was moved or renamed, and will want to reestablish a link to it. However, you should always close the document in Word before renaming it.

Can I use AuthorProof with the cloud?

Yes, you can use it with OneDrive (Microsoft), iCloud Drive, Dropbox, etc. The first time you start tracking a document that is in a cloud-synced folder, the program will let you know that the archive for this document needs to be set to “Always keep on this device.” This tells your computer to keep the archive folder locally (on your computer) rather than on the cloud.

Should I continue tracking my Word doc across many iterations?

This is your judgement call. Me personally, I think I would use one archive (Word doc) for the first draft of my novel, then a second one for the second draft, etc. If you had 50 drafts of a novel spread out over ten years, this might make your file unwieldy.

A note about the alternatives to AuthorProof

Google Docs version history, OneDrive version history:

I don’t have extensive experience with Google’s version history, but there are some reasons why I personally do not use it. I’m a professional writer, which means I have to work in Microsoft Word at some point. My concern with Google is privacy. Both Microsoft and Google kinda sorta say that they do not use your content to train AI. Google says: “We do not use your Workspace data to train or improve the underlying generative AI and large language models that power Gemini, Search, and other systems outside of Workspace without permission.” What does “without permission” mean? Where is the permission granted? Is it buried in 34 pages of terms of service? On the same page: “If you’re a Workspace user with a personal account and you choose to share data, including your Workspace data, with Gemini Apps or Search services for Personal Intelligence, or through screen actions (including screenshots), this data will be processed according to their respective terms and policies, and may be used for model training and improvement.” My decision to stay with Microsoft is to limit my exposure to one AI company, not two. While the question about training is a substantial one, the only Microsoft response to this I could find publicly is a Q&A one one of their Community Forums where I also didn’t really like Microsoft’s answer: they said they don’t train and that this is in their Terms of Service, but would not answer when several people asked them to point out where. They have verbally said they don’t use customers’ content to train, but I think we’d all like a bit more than that.

With respect to Microsoft OneDrive, its version history thing was not built for the purpose we are talking about—it was built to save content that you accidentally deleted. The history only goes back 30 days or 25 versions. My understanding—and someone correct me if I’m wrong—is that Google versions can go all the way back to the start. So this would give you the equivalent of a filing cabinet of all the versions of your work—I suppose you could send particular days—maybe even send all of them if whoever is asking is interested in looking through them. (My application produces one file you can look through that documents all changes across time. It’s ugly to look at, but that’s what it does.) The one issue with Google version history is that you can manually delete versions—and maybe that is the version where 40,000 words magically appeared overnight. It would be a bit hard to explain why you did that though.

Scrivener

Scrivener has a snapshots function that I imagine was built to protect authors from going too hogwild when editing. You snap a version of where you are right now, and then all those are logged in case you want to go back to previous text. This is all nested within Scrivener’s sections. Video on how to do this here. This is a nice feature but I’m not sure how portable it is. As in, you get a window that has whatever version you want in it— I suppose you could send the entire Scrivener file to a third party for them to look through. You can also compare old text via new text via a track changes like highlighting. So this would rely on that third party who may or may not trust you paging through your many different versions. However, you can also manually delete snapshots, which is a problem.

“date created” and emailing yourself/ someone else

Microsoft Word and other programs has meta data that denotes when the document was originally created and last modified. Which is great for those of us who are honest. Unfortunately, there are programs where you can manually alter that metadata. Many writers regularly email themselves their project—it used to be that this was in case you had a catastrophic computer failure, not to prove that you wrote something! You could also have emails to a trusted third party, like a literary agent or critique group. It is possible to forward an email you sent in the past and manually change the date it was ostensibly sent, or even change the content of the attachment. However, you should be able to have the receiver of your emails verify that you did in fact send such and such document on such and such date, and what the content of that document was. All of this BS ultimately just highlights the importance of being part of a writing community.

Other programs

There is a program called Ellipsus that has some form of author certification in its paid version. I can’t vouch for it because I haven’t tried it. There are a couple of foreign programs being developed- I will update this space and will very likely be writing on Substack about them as I come to know more information.

Common sense.

Real writers—writers who are not trying to pass of AI writing as their own—can have in depth and intelligent discussions about where their ideas come from, who is their work in conversation with, why they made particular editorial decisions. The app that I created is quite possibly overkill, but in all honesty when I designed it I was imagining and author who had been falsely accused and there was a lawsuit; right now if the standard of “proving” that something was written with AI was running it through Pangram or observing that the writing is overly flowery and sometimes has bad metaphors, that would be deeply concerning to me. I think it behooves authors to keep good records and to have an AI statement somewhere on their website to make it clear and and unambiguous what role, if any, AI played in their writing.