A new study introducing the CASPER framework reveals that AI writing models consistently strip away narrative mystery, relying on safe archetypes and tidy resolutions, proving that scaling up model sizes fails to improve character depth or capture genuine human contradiction.