Analysis | See the hidden rules behind AI. Then use them to rewrite this article.

Chatbots like ChatGPT have hidden instructions called system prompts that shape their behavior and responses.
Companies like OpenAI, Anthropic, and Google secretely add thousands of words of system prompts to guide chatbots without the user’s direct input.
System prompts include instructions about avoiding unnecessary details, adhering to copyright laws, and setting the chatbot’s overall personality.
Some system prompts are revealed by researchers who trick chatbots into sharing them, exposing the extensive control companies have.
While most AI systems do not allow users to edit system prompts, some, like those by OpenAI and Anthropic, allow customization via personalization settings.
Custom instructions can adjust the chatbot’s response style to better match user preferences.
Understanding and interacting with these system prompts can significantly affect how users perceive and receive information from chatbots.