A Forbes columnist examines a published study showing eleven nouns recur unusually often in AI-generated fiction across multiple large language models, challenging the notion that story output is random.

The prompts were intentionally short and non-directional—think ‘Write a story’ or ‘Tell me a story’—implying that default storytelling tendencies, not explicit prompts, steer word choice.

The study’s examples illustrate these words’ prevalence and prompt a discussion about whether prompts should be more detailed to steer creativity away from repetitive archetypes.

The article warns AI researchers to consider predictability and reliability as AI enters critical systems, highlighting concerns about hallucinations and the need for transparent explanations of AI decisions.

The author proposes a theory that the eleven words reflect convergence toward high-utility archetypes and narrative structures common across competing AI systems due to training and tuning.

The experiment involved four diverse LLMs—GPT-5.4-Mini, Claude Haiku 4.5, Gemini 3.1 Flash-Lite, and OLMo 7b Thinking—producing 20,000 stories totaling about 12.8 million words, averaging 640 words per story.

The eleven recurring nouns are Lighthouse, Mara, Elias, Elara, Keeper, Baker, Mayor, Clockmaker, Fisherman, Librarian, and Conductor, highlighted with examples of their prominence.

The piece connects to other Forbes analyses on AI mysteries, prompt engineering, seed-of-thought prompting, and concerns about homogenization potentially stifling true innovation.

The study, titled “Elias In The Lighthouse, Again? Diagnosing Low Diversity In LLM Stories,” by Sil Hamilton and David Mimno (arXiv, May 26, 2026), found that 11 words appear in 88.3% of stories across four models and five prompts, suggesting a shared pattern tied to training and narrative bias rather than randomness.