AI answers differently depending on how a prompt is worded; does that same sensitivity show up in the fiction it writes?

This explores whether AI's sensitivity to framing, both in how it reacts to the way a prompt is worded and in how it positions itself rhetorically, carries over from expository and scientific writing into creative work. The short answer is that the collection shows the pattern does carry over, but in an odd form. AI is very sensitive to framing on the way in and fairly flat in its own rhetoric on the way out, and that imbalance looks the same in fiction as it does in argument.

Start with the input side. Models respond to framing that adds no new information. Adding a line like "This is very important to my career" to a prompt consistently improves performance across several models, and positive emotional words account for more than half of the gains Can emotional phrases in prompts improve language model performance?. So models clearly pick up on motivational and rhetorical cues. A useful way to think about this is the old split between logic, credibility, and emotional appeal, which has been used to show that every AI explanation pulls all three levers at once, whether the designers meant it to or not How do logos, ethos, and pathos shape AI explanations?.

The output side is where it gets interesting. In expository prose, LLMs have mastered grammar and structure but avoid taking an evaluative stance. They reach for neutral, descriptive wording where human writers choose words that carry judgment and evidential weight, so the prose is well organized but makes no real argument Why does AI writing sound generic despite being grammatically correct?. The fiction research finds the same habit in creative form. AI stories over-explain their themes, prefer tidy single-track plots, and avoid moral ambiguity, while human writers use nonlinear time and leave things unresolved. This held across all five major models tested Do AI stories explain their themes more than human stories do?. Over-explaining a theme is the creative version of refusing to take a stance: the model spells out the meaning instead of trusting the reader to feel the tension. These choices also run deep. AI fiction can be identified with 93% accuracy from narrative structure alone, with no stylistic cues, because fixing it would take rewrites rather than surface edits Can AI stories be detected without analyzing writing style?.

There's a theory in the collection for why this happens in both domains. Human writing contains a built-in appeal to the reader's attention, and AI text lacks it, which explains the "aloofness" readers report Does AI writing lack the internal appeal to attention that humans use?. A related argument holds that AI output is a kind of residue carrying the markers of communication without an actual speaker addressing anyone, so readers fill in the missing intent themselves Does AI generate genuine utterances or just text patterns?. If either account is right, the gap isn't specific to genre. A story, like an argument, depends on someone deciding what to withhold from a particular reader, and that's exactly the move AI doesn't make. Framing still reaches people in creative and personal writing too, just in a different way: AI writing assistance shifted readers' impressions of the writer on all 29 measured dimensions, toward more confidence, extremity, and agreeableness Does AI writing assistance change how readers perceive the writer?.

One gap to flag directly: the collection has no study that tests whether reframing a *creative* prompt (emotional stakes, a persona, a stated audience) changes the quality of fiction the way EmotionPrompt changes task performance. So the link here is an inference across the papers, not a measured result. The takeaway you may not have expected is that AI's sensitivity to framing runs one way: it is easily moved by how it's addressed, yet in both arguments and stories it has trouble addressing its own reader.