AI Fiction in the Wild Visualizer
Infinite Stories and Common User Types
AI Fiction in the Wild Visualizer
Infinite Stories and Common User Types
How are people using AI models to generate fiction and stories?
These are ~193,000 real-world, anonymized ChatGPT-user conversations that were collected with users' consent by researchers at the Allen Institute for AI between 2023 and 2024 (see WildChat). We used a model to classify "fiction"-related conversations in English, including requests for stories, novels, and scripts, but also roleplay, hypothetical scenarios, and more.
We organize conversations by estimated user to show common patterns. Learn more in our paper →
Content warning: This dataset contains conversations that are explicit, offensive, and sexually graphic. You can hide content that has been tagged explicit or toxic. Be aware that other offensive material may not be caught by these filters.
A note on user demographics and motivations
Who are the users in this dataset, and why did they use WildChat? We don't know.
The chatbots were hosted on Hugging Face—a community for sharing machine learning models and datasets. You can see the interface users interacted with, which is different from ChatGPT's interface.
Because the models were hosted on HuggingFace, we might guess, as the original creators do, that users skew more technically literate and "online" than the average ChatGPT user.
The WildChat instances of ChatGPT were also free, did not require a login or account, and did not have rate limits. This might mean that users are from lower-income backgrounds, from countries where ChatGPT is banned, or that they were more invested in pushing the boundaries of the model with explicit or prohibited content (not connected to a real account).
WildChat is not a representative sample of all ChatGPT users. But it's a rare window into real user behavior.
Some users send ChatGPT the same creative prompt — again and again, with small tweaks each time.
This user sent 15 variations of the same Stranger Things scene over one week. Green = added, red = removed vs. previous prompt.
Explore this cluster →The same user also explores many different franchises with a similar pattern — "evil reveal" or "married" scenes across anime, games, and TV.
These 5 prompts span just 3 days — each a different fictional universe.
Explore all 407 conversations →