On AI, Authorship, and Ethical Transparency in Creative Works
So, I’d like to put on my scholarly hat for a moment and have an educational discussion on AI, authorship, and transparency as it relates to fanfiction communities.
Lately I’ve seen discussions about the distinction between “evaluative” and “generative” AI use in writing, and I think it’s worth clarifying some points for the sake of better community understanding. These topics matter not because anyone is looking to shame writers, but because there are well-established boundaries in publishing, academia, and digital ethics that affect how we define authorship.
Below are a list of items that I believe are important to address:
1. AI Cannot “Learn Your Voice” Without Generative Use
A recurring claim I’ve seen is that AI is used strictly for evaluation, yet the same users say the model has “learned their voice” over time.
AI cannot “learn your voice” through passive evaluation. Research consistently shows that large language models do not adapt to individual user style or preferences through ordinary usage. Model behavior only shifts through deliberate alignment techniques such as fine-tuning, reinforcement learning, or controlled inference methods (Li et al., 2023). In other words, simply having an AI “evaluate” your writing does not cause it to internalize your voice, and passive exposure does not produce learning in these systems.
The only way an AI adapts to your prose rhythm, tone, and metaphor patterns is by generating content based on your inputs, meaning that the user is engaging in generative refinement, and not mere proofreading.
If an AI is shaping rhythm, pacing, flow, metaphor structure, or tonal consistency, that is authorship influence and no longer gets to be defined as editing.
2. Revision is Not Ethically Neutral When the Tool Generates Output
There’s a misconception that “I wrote the first draft, so the rest doesn’t count.”
But in academic and publishing ethics, if a tool meaningfully alters style, structure, or expression, the tool is considered a co-authoring agent. Even replacing sentences, restructuring paragraphs, or smoothing dialogue can constitute generative contribution.
Multiple professional organizations make a clear distinction between editing tools and generative systems, and all agree that when a tool contributes to the expressive elements of a text, that contribution constitutes authorship. COPE’s 2023 position statement emphasizes that AI systems generate content and therefore cannot be treated as neutral editors; they produce text that meets authorship criteria even though they cannot assume responsibility for it (Committee on Publication Ethics, 2023). Similarly, the Modern Language Association (2023) states that any text that is generated or significantly shaped by AI must be formally acknowledged, as failing to do so misrepresents the true source of authorship. The U.S. Copyright Office reaches the same conclusion from a legal standpoint: when an AI tool determines or materially contributes to stylistic, structural, or expressive choices, that portion of the work is not human-authored and cannot be claimed as such (U.S. Copyright Office, 2023). Taken together, these guidelines establish that meaningful stylistic or structural alteration by AI moves the tool’s role from editing into co-authoring, requiring transparency.
3. Intent Alone Cannot Establish Authorship
Another idea circulating is: “If the intent comes from me, the work is mine.”
Unfortunately, that is not how authorship is defined in any formal discipline. Authorship is tied to expression, and not intent (U.S. Copyright Office, 2023).
If the expression is significantly shaped by a non-human system, intent doesn’t override the system’s contribution.
If intent alone determined authorship, ghostwriting, translation AI, and text generators would all be considered ethically equivalent to self-written work, which they are not.
4. The Editing vs. Generating Distinction Is Often Misunderstood
Many claim, “I don’t tell the AI to write scenes, so I’m not generating.”
But generative AI does not require scene-level prompts to be generative.
Examples of generative influence include: restructuring sentences, rewriting passages for tone, creating metaphor suggestions, smoothing dialogue, adjusting prose rhythm, rewriting transitions.
Each of these is generative because the new text did not originate from the author.
If someone rejects “robotic” suggestions yet continues to use the tool until the flow feels natural, the tool is shaping style. This is very different from grammar correcting.
Ultimately, this post is not meant to be anti AI. AI has many valuable uses, but when it begins shaping expression in creative writing, its role becomes an ethical consideration rather than a simple tool choice.
Committee on Publication Ethics. (2023, February 13). Authorship and AI tools: COPE position statement. https://publicationethics.org/guidance/cope-position/authorship-and-ai-tools
Li, Y., Wei, F., Zhao, J., Zhang, C., & Zhang, H. (2023). RAIN: Your language models can align themselves without fine-tuning. arXiv. https://arxiv.org/abs/2309.07124
Modern Language Association. (2023). MLA guidance on the use of generative AI in academic writing. https://style.mla.org/citing-generative-ai/
U.S. Copyright Office. (2023, March 16). Copyright registration guidance: Works containing material generated by artificial intelligence (Policy Statement). https://www.copyright.gov/ai/ai_policy_guidance.pdf
Author is a bilingual person who is one class away from a whole ass science degree, capable of reading primary sources.
gummygunk
blakekathryn
snowghoul
ellievsbear
yousharknotpass
discoholicmusic
pixelskylines
stormborns
romaincrisis
