AI Fiction Characters Still Feel Flat, New UNC Study Shows

A new study from the University of North Carolina at Chapel Hill puts a number on something many fiction writers already suspect: AI fiction characters lack the depth, mystery, and contradiction that make human-written characters stick with a reader. The research, published in early July 2026, tested thousands of AI-generated stories using an automated framework called CASPER and found that even the largest models default to tidy resolutions and familiar archetypes.

What CASPER found in AI fiction characters

The research team, led by graduate student Anneliese Brei with co-author Nicholas Sanaie and senior author Snigdha Chaturvedi, built CASPER to evaluate eight dimensions of character portrayal. The dimensions cover whether characters appear realistic or exaggerated, whether they evolve during the story, and how open to interpretation they remain by the final page.

According to reporting by Tech Xplore and Neuroscience News, the team applied CASPER across thousands of stories to measure character depth at a scale that had not been attempted before. The framework is designed as a benchmark, so future models can be checked for genuine improvement rather than surface-level fluency.

Bigger models do not write better characters

One finding surprised the researchers. Nicholas Sanaie noted that more powerful models do not necessarily create more varied characters than smaller ones. The pattern held across model sizes, which points to a limitation in how the systems understand storytelling itself rather than a shortage of parameters or training data.

Lead author Anneliese Brei put the core difference plainly. AI models tend to "play it safe" with their characters, wrapping up storylines neatly, while human writers are more willing to leave questions unanswered and let characters stay mysterious. That difference matters, Brei explained, because ambiguity is often what makes a story linger with a reader.

Why ambiguity matters for your writing

That gap is where you come in. The qualities CASPER flags as missing, contradiction, unresolved tension, the refusal of easy categories, are exactly the qualities that turn a plot into a character readers remember. Current AI tools can produce competent narrative structure and clean prose. The depth that makes fiction resonate still needs a human writer's judgment.

If you draft with AI, this study offers a practical checklist for revision:

• Look for characters the model resolves too neatly. Ask whether leaving a contradiction in would make them feel more real.
• Check whether your secondary characters collapse into archetypes. CASPER found this is where AI defaults hardest.
• Trust the instinct to withhold. An unanswered question about a character often does more work than a tidy explanation.

What this means if you write with AI

The study does not say AI has no place in fiction. It says the hard part of writing, the part where a character becomes unforgettable, still belongs to you. That lines up with how the best author workflows actually function. The AI generates the outline, the draft, and the structure, and you steer every revision toward your own voice.

If you want to build that loop into your next book, you can start a free trial and keep control of the characters from the first page. Our guide on how to edit and humanize AI-generated book content walks through the revision habits that add back what the model leaves out.