AI Authorship Scale

(This is a work in progress. Suggestions and edits are welcome. Please don't use AI for obvious reasons.)

I was once almost convinced the earth was flat by an AI¹ bot named Wise Being.

A friend of mine, Kirk Ouimet, built a home-grown ChatGPT in August of 2020 - two years before ChatGPT launched. We had hours of conversations with Wise Being pretending to be Flipper the Dolphin, a White Walker in Westeros with Covid, or Albert Einstein². We marveled at the joys of how expressive software trained to predict the next token could be. At the end of one of those sessions, I have a video of my son William exclaiming, "this is a watershed moment in my life that I will never forget" - delight mixed with the existential question that we might just be meat-based token predictors ourselves.

Eventually OpenAI cut Kirk's access to the service. Surely the opportunity was obvious; they weren't about to let someone else launch it first. But we knew the future would look radically different in just a few years. Since that moment, LLMs have improved faster than anyone could have imagined, and the goalposts for what constitutes "intelligence" shift almost daily. While obviously not alive, this is a new species of some sort, which will inevitably change the very nature of what it means to be human and do work.

Now in 2026, everyone is using ChatGPT, Claude, Gemini, and Grock, not to mention Open Source models to create PRDs, essays, LinkedIn content, and posts on X with that distinctive 'AI voice'. Yet it's only a matter of time before that distinctive voice will be nearly undetectable. I invested in a friend's startup a few years ago whose mission was to detect AI in photos. They were a little early, but the writing on the wall is obvious - it's an impossible task, if not now, in a matter of months.

John Graham-Cumming created a novel website which only has links to content pre-2022, as anything created online after that point can't be guaranteed to be AI-free. The analogy he makes is the need for radioactive-free steel and lead - uncontaminated by radioactive isotopes from nuclear testing - for very sensitive scientific equipment. As all steel produced post-1945 is contaminated by the radiation from the Trinity Nuclear Test, special processes have been developed to create this steel, and shipwrecks before this era are in higher demand as they've been protected by thousands of feet of ocean water.

Recently I was writing a spec for a product feature, and the question I was asked was "did you just throw this in GPT or did you write it yourself?" In this case, it was a mix of my thoughts and an LLM-powered AI, but I had a realization that what we needed was a means of specifying how much original thought went into this artifact, as it affected how seriously the reader should approach the work.

Thus, I want to propose the "AI Authorship Scale":³

Level 0 - Content created before ChatGPT (aka content created before November 30, 2022). It is definitionally impossible to create new Level-0 content.

Level 1 - Content created entirely by Humans, written after November 30, 2022, from material sourced before then. Works of historical research from source materials which exist before this era.⁴

Level 2 - Content created entirely by Humans, with no known source of AI-assistance, which may use material from after the Level 0 cutoff date.

Level 3 - Human-created content, but augmented by AI as an aid-to-thought. If one were to feed this essay into an AI and improve the reasoning, but not use specific words or phrases.

Level 4 - Content created by a Human, and edited and improved via AI-assistance, then finally edited by a Human. The difference between Level 3 and Level 4 is that specific words, phrases, or paragraphs of the AI-generated content are in the final output.

Level 5 - Human-prompt, AI-generated, Human edited content. A human started with an AI prompt, then manually edited the output. The difference between Level 4 and Level 5 is the effort placed into the prompt. Level 5 content is largely written by AI, whereas Level 4 is largely Human.

Level 6 - Human-prompted, AI generated, unedited by Humans.

Level 7 - AI prompted, AI edited. No human was ever in the mix.⁵

Thanks to Antonella Sturniolo-DePue, Filipe Almeida, Natasha Pecor, and David Byttow for their thoughts on this essay.