Why chatbots keep reusing ‘Elias Thorne’ in stories
ChatGPT, Claude, and Gemini often reuse names like “Elias Thorne” and roles such as lighthouse keeper, clockmaker, or explorer when asked to write short stories. Researchers generated 20,000 stories from the three chatbots using five prompts.
They found that 11 repeated words, including names and job roles, appeared in more than 88% of the stories. The likely causes include safety tuning, which pushes models toward safe and familiar outputs, and training data that includes text made by other chatbots.
The researchers point to GPT-3.5 as one important early source because its outputs helped form public datasets that later influenced other datasets and models. The same repeated story patterns are also showing up in AI-made books on Amazon, low-quality YouTube content, and fake news sites.
Key points
- ChatGPT, Claude, and Gemini can repeat similar names and roles in generated stories.
- Researchers compared 20,000 chatbot-written stories made from five prompts.
- A small set of 11 words appeared in more than 88% of the stories.
- Safety tuning and reused training data may be making different models sound more alike.
- Makers should edit AI drafts for originality before using them in public content.
Quick term guide
- Lighthouse
- A tool that tests how well a website performs and follows best practices.
- chatbots
- Programs that answer people through a chat-style conversation.
- safety tuning
- A process that trains a chatbot to avoid harmful or risky answers.
- training data
- The collection of information used to teach an AI how to recognize patterns and answer questions.
- datasets
- Collections of data used to train or test a model.
- Solo makers
- People who build and launch their own products or services entirely on their own.
- marketing
- The activities used to tell people about a product and encourage them to buy it.
- audience
- An audience means the people who regularly see or follow your posts.