How to prove you wrote it — not an AI.

The problem

Writers — novelists, students, journalists, copywriters, ghostwriters, academics — are now routinely accused of having used AI to produce text. The accusation arrives whether or not it is true. Editors, professors, publishers, and platforms all run AI-detection tools, and those tools are wrong often enough that the accusation often arrives at honest people first.

A better detector is not coming. Detection tools hallucinate at scale and contradict each other. The fix is to make the argument about when the writing existed, not whether it looks AI-generated. If a draft of the same text demonstrably existed on disk before the supposed AI prompt, the detection question becomes moot.

What you need

Nothing to install. A browser. A file. The file can be your manuscript draft, a research note, a chapter outline, an email, a screenshot of your editing window, a recorded voice memo, a Word document — any byte sequence. The instrument treats everything as raw bytes; it does not read your prose.

What you do

• Save your draft as a file on your machine. Date the filename if you want — the date is for you; the proof does not depend on it.
• Drop the file at orphograph.com. The browser computes a SHA-256 fingerprint locally; the file's contents never leave your machine. Only the 64-character fingerprint is submitted.
• Save the receipt. You receive a 16-character receipt ID, a public URL of the form orphograph.com/r/<id>, and five small .ots proof files. Keep them next to the manuscript in your normal backup folder.

That is the whole inoculation. The receipt's fingerprint is committed to a Bitcoin block within about an hour. From then on, the receipt is evidence that the file existed by no later than that block's mined timestamp.

What the receipt actually proves

It proves one specific, narrow fact: a file with this exact SHA-256 fingerprint existed on or before Bitcoin block N.

That is enough for the AI-suspicion conversation, because the workflow becomes:

• You are accused on, say, October 14.
• You produce the receipt showing the file existed on, say, June 3.
• Anyone with the original file can re-hash it and confirm the receipt matches.
• Anyone with a Bitcoin node can verify the block was mined June 3.
• The accuser is now arguing you predicted on June 3 exactly which AI-generated text would later best match a future accusation, which is absurd.

Three workflows for three kinds of writers

Novelists and long-form authors

Anchor the draft at the end of each writing day. A weekly or chapter-end cadence is fine if daily is too much friction. Keep the receipts in the same folder as the manuscript backups. Five years from now, when a "you used AI to write that" question arrives — even casually on social media, not through a formal challenge — you have a stack of timestamped artifacts tracing the actual writing process from the inside.

Students writing essays

Anchor the file when you first finish a draft, and again when you submit. If the assignment is contested later, the receipt set tells the story: here is the draft from Wednesday night, here is the version submitted Thursday morning, here is what the rubric was on Tuesday. Two anchors. Free tier.

Journalists and academics

Anchor source documents — recordings, leaked PDFs, datasets — the moment you receive them. Anchor your own manuscripts at finalization. The receipt set forms a paper trail that a fact-checker, editor, or court can verify without trusting you or the publication.

What it cannot do

• It cannot prove you are the author. Anyone with the file can anchor it. Authorship is established through other means — handwritten notes, document version history in editing software, prior correspondence, witnesses. The timestamp is necessary, not sufficient.
• It cannot prove the file was unedited between draft and submission. Each anchor pins one specific byte sequence. Edit one comma, and the fingerprint changes — that is a different file with its own receipt needed.
• It cannot prove the prose is human-written. Nothing can prove that. But it can prove the prose existed before the AI tool being blamed was capable of generating it — which is usually the underlying question.

What survives the service

Every receipt this office issues can be verified later through OpenTimestamps, the open public standard the office uses underneath, against any copy of the Bitcoin chain. If the office closes tomorrow, the receipts you hold today still work. The instrument is in your hands, not on a server you do not control.

Try it

Open orphograph.com. Drop a draft. Save the receipt. You now have a defensible date of record for that file. If the AI-suspicion conversation ever finds you, the receipt is what you hand over.