Helpful Semantics for Talking about "AI Writing"

A list of helpful generative AI category definitions to easily reference what's under an author's hood.

The dichotomy between “AI writing” and “human writing” is fake. Only people who haven’t spent at least an hour or two playing with an LLM think this way. They say things like, “If you use AI, you’re not a real writer,” or, “Using AI will destroy your writing.”

These harsh proclamations are totally disconnected from the millions of subtly different ways a writer may use Large Language Models in their craft. AI is highly flexible and responds to specific inquiries. It isn’t like using a set program from proofreading, or even something like Grammarly, which merely generates suggestions for improving your prose. AI does, or attempts to do, whatever you prompt it to do. Its flexibility resists any kind of dichotomy imposed on it, and it is no wonder that writers who have used LLMs are up in arms about being told they aren’t “real writers.”

The truth is that your authenticity as a writer lies on a spectrum. The more you depend on AI, the less of a writer you are, having surrendered more and more of your craft to the machine, and leaving less and less for your own voice and perspective. If AI is generating entire texts for you, you are a facilitator of writing that isn’t yours, equivalent to a teacher telling a student to write an essay on X, Y, Z. It’s AI writing, and your simple prompt is facilitation, not authorship. On the other hand, if your text is completely your own but needs a quick polish, a “second set of eyes,” you’re AI-edited, which isn’t functionally different from being edited by another human.

Specific things needs specific terms, because without them we are all fumbling around in the dark. I submit to you that we need a standardized glossary of categories for understanding broadly how writers employ this technology. You may find that you fit in multiple categories. That’s okay. It’s a handy reference for understanding non-fiction primarily. Fiction is a different beast, and I may need an addendum or other glossary to treat it adequately.

Even if you find this glossary insufficient, I beg you to be precise with your critiques. Don’t write comments like, “this is AI writing.” Say something like, “this essay sounds like the writer got some help from an LLM,” or “it bears the hallmarks of Chat GPT,” or “I’m struggling to connect with this; it feels inauthentic.” Nobody benefits from broad-brushing except people who thrive on outrage.

My hope is that writers can use these categories to talk openly about how they employ AI, because transparency does equate to some amount of goodwill and honesty is authenticity at its purest.

Categories of AI Use in Writing

AI Writing or AI text

Text generated by an LLM with little to no editing by a human author.

AI-Moderated

Text created with heavy inputs from LLMs, but its primary notional framework originated with the human writer. Whole clauses, sentences, or paragraphs may be AI-generated, but the overall content is closely moderated by the human writer. Text may sound generic or “AI” to varying degrees, but thoughts demonstrate human perspective and reflect the writer’s intentions.

AI-assisted

Thoughts and ideas are created primarily by the human author, with LLM assistance in articulating or re-articulating the original thoughts line by line or paragraph by paragraph, and fleshing out minor points. As with AI moderated text, the prose may sound AI-generic, but the underlying thoughts

AI-brainstormed

Text is written by a human but LLMs were used in the brainstorming process to refine ideas or conduct research.

AI-edited

Text is written by a human but passed through an LLM strictly for line editing, including proofreading. Suggestions are adopted in bulk with little moderation.

AI-assisted editing

Text is written by a human but LLMs are used in the editing process for generating or brainstorming line edits. Final edits are made by the human author/editor.

Your suggestions for other categories or revisions to categories and terms are welcome.