AI has many wonderful capabilities, and I have no doubt that only a fraction of them have yet to be discovered. Universally, however, we innately distrust art or craft that’s made by a machine. There needs to be in each some semblance of the human spirit—whether we’re reading an article, listening to a song, or looking at a painting, we want to know that whoever, or whatever, is on the other side of that creation knew the same highs and lows of existence as we do, otherwise there is something within us, perhaps our soul, that prevents us from forming a bond with that creation, or be moved by it.
In steps Ink & Proof to reassure readers that the stamped piece of writing they’re about to read is verifiably from a human, and to let them know that if AI was used, to what capacity. The four tiers below exist with their own easily-identifiable stamp, instantly passing along to a reader the knowledge of if and how AI was used in the creation of the piece they’re about to read.
The Four Tiers
Written by Hand
Composed in the Ink & Proof editor. No AI assistance used during drafting, research, or editing. (Note: The lack of AI use in research is self-declared by the author, as it cannot be reasonably observed by an application, or could otherwise be circumvented in a multitude of ways.)
Ex: A personal essay drawn entirely from memory and lived experience, researched through books and interviews with no AI involvement.
Written by Hand, AI-Assisted Research
Composed in the editor. AI was used for research, fact-checking, or source-gathering; no AI-generated text appears in the final work.
Ex: An analysis of Fed policy, where a writer used AI to summarize recent FOMC transcripts, but wrote every sentence themselves.
Written by Hand, AI-Assisted Editing
Composed in the editor. AI was used for copyediting, grammar, or structural feedback on the author's own draft; no AI-generated text appears in the final work.
Ex: A finished draft run through AI to catch grammatical inconsistencies and tighten passive constructions, but the ideas and prose are entirely the author's.
Human-Directed
AI was used substantially in drafting. The author made meaningful editorial choices throughout, directed the structure and argument, and takes full authorial responsibility for the finished work.
Ex: A writer prompted their LLM of choice to draft three sections on a topic they know well, rewrote substantially for voice and argument, cut two sections entirely, added an original conclusion.
What Ink & Proof Is Not
What Ink & Proof is not is a 100% guarantee of accuracy in AI use assessment. We’ve taken every reasonable measure to monitor a writer’s work in the hopes of passing along an honest report of their AI use to the reader, but we fully acknowledge that there are loopholes that could be leveraged by a determined writer that could disguise, even to us, the extent of their use of AI-generated content. We take this misuse seriously, and make one of our primary focuses ensuring that this inauthentic use of our product is as inconvenient and disruptive to a writer’s workflow as possible. To help discourage this kind of misrepresentation, writers that utilize Ink & Proof, in addition to writing in our monitored editor, must verify their identity, and are notified upon registration that if they’re found to be circumventing our policy and terms of use they’ll have their previous stamps of approval tagged and be publicly blacklisted from using our products in the future.
The stamp isn’t “proof of no AI.” It’s friction plus a public commitment. Friction matters because most AI “cheating” in writing isn’t adversarial—it’s lazy. The median writer cobbling their Substack post together in 20 minutes with ChatGPT isn’t going to retype 2,000 words of AI output across three sessions over two days with realistic edit density just to earn a badge. The cost of cheating exceeds the cost of just writing.
What We Observe
- Session behavior — Number of distinct writing sessions, duration of each session, total cumulative writing time across all sessions.
- Word count evolution — Word count at the start and end of each session, growth trajectory over time—did the piece build gradually, or appear fully-formed?
- Edit density — Characters added vs. deleted across the draft's lifetime, ratio of deletions to final word count. Real writing involves meaningful revision; pasted text doesn't.
- Paste events — Size of each paste action in characters, whether any single paste exceeds 20% of the final word count (flagged as anomalous).
- Snapshots — Full content captures every 5 minutes per session, stored as an audit trail.
Behavioral Thresholds
The following thresholds apply to the top three tiers and are published openly:
- ◆Minimum 2 distinct sessions
- ◆At least 10 minutes of writing time per 500 words of output
- ◆Edit density ≥ 15% (deletions ÷ final word count)
- ◆No single paste event > 20% of final word count
What it does not observe: anything outside of the editor—browser tabs, external tools, voice dictation apps, or a writer re-typing AI-generated text by hand.
Everything above is supplemented by the author’s self-declaration via the disclosure dialog, and the resulting proof object is cryptographically signed so the record can’t be altered after the fact.
The Threat Model
What kinds of cheating the system defeats and what kinds it doesn’t.
Will Catch
- Bulk-pasting of AI-generated text into the editor
- Single-session, no-revision “dumps” (fails session count and edit density checks)
- Fast, unrevised drafts relative to word count (fails time-per-word threshold)
Will Not Catch
- Retyping AI output manually (passes all behavioral checks)
- AI research conducted in another tab, then written from memory
- Light AI assistance the author simply doesn’t disclose
The system defeats laziness and carelessness. It doesn’t defeat deliberate, patient deception.
Why We Have a Human-Directed Tier
Ink & Proof is not anti-AI, and we acknowledge that in certain circumstances AI-generated output (with human review) might be useful—providing simple instructions to include in a product guide, succinctly summarizing something complex, and so on. We’d be dismayed to find out that our favorite novel was written by AI; the instructions to our TV remote? We likely wouldn’t care.
We don’t want to exclude or vilify this kind of work. We simply want the reader to be fully aware that AI was leveraged in the actual crafting of the text they’re about to read.
A Commitment
When we update the methodology, we version it publicly, and old Proofs link to the methodology version they were certified under.
We remain committed to the integrity of every stamp—continually exploring new verification methods and writer monitoring tools to ensure that an Ink & Proof stamp carries the strongest attestation that technology and honest methodology can reasonably provide.