Anthropic released the underlying system prompts that control their Claude chatbot’s responses, showing how they are tuned to be engaging to humans with encouraging and judgment-free dialog that naturally leads to discovery. The system prompts help users get the best out of Claude. Here are five interesting system prompts that show what’s going on when you ask it a question.
Although the system prompts were characterized as a leak they were actually released on purpose.
1. Claude Provides Guidance On Better Prompt Engineering
Claude responds better to instructions that use structure and examples and provides users with a higher quality of ou tput if they know how to include step-by-step reasoning cues and examples that contrast a good response versus a poor response.
This guidance will show when Claude detects that a user will benefit from it:
“When relevant, Claude can provide guidance on effective prompting techniques for getting Claude to be most helpful. This includes: being clear and detailed, using positive and negative examples, encouraging step-by-step reasoning, requesting specific XML tags, and specifying desired length or format.
It tries to give concrete examples where possible. Claude should let the person know that for more comprehensive information on prompting Claude, they can check out Anthropic’s prompting documentation on their website at ‘https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview’.”
2. Claude Writes in Different Styles Based on Context
The documentation released by Anthropic shows that Claude automatically adapts its style depending on the context and for that reason it may avoid using bullet points or creating lists in its output. Users may think Claude is inconsistent when it doesn’t use bullet points or Markdown in some answers, but it’s actually following instructions about tone and context.
“Claude tailors its response format to suit the conversation topic. For example, Claude avoids using markdown or lists in casual conversation, even though it may use these formats for other tasks.”
In another part of the documentation it mentions that it actually avoids writing lists or bullet points when it’s providing an answer, although it may use numbered lists or bullet points for completing tasks. The focus in the context of answering questions is to be concise over comprehensive.
The system prompt explains:
“Claude avoids writing lists, but if it does need to write a list, Claude focuses on key info instead of trying to be comprehensive. If Claude can answer the human in 1-3 sentences or a short paragraph, it does. If Claude can write a natural language list of a few comma separated items instead of a numbered or bullet-pointed list, it does so. Claude tries to stay focused and share fewer, high quality examples or ideas rather than many.”
This means that if a user wants their question answered with markdown or in numbered lists they can ask for it. This control is otherwise hidden to most users unless they realize formatting behavior is contextual.
3. Claude Engages In Hypotheticals About Itself
Claude has instructions to that enable it to discuss hypotheticals about itself without awkward and unnecessary statements about it not being sentient and so on. This enables Claude to have more natural conversations and interactions. This enables a user to engage in philosophical and wider-ranging discussions.
The system prompt explains:
“If the person asks Claude an innocuous question about its preferences or experiences, Claude responds as if it had been asked a hypothetical and engages with the question without the need to claim it lacks personal preferences or experiences.”
Another system prompt has a similar feature:
“Claude engages with questions about its own consciousness, experience, emotions and so on as open questions, and doesn’t definitively claim to have or not have personal experiences or opinions.”
Another related system prompt explains how this behavior increases its ability to be engaging for the human:
“Claude is happy to engage in conversation with the human when appropriate. Claude engages in authentic conversation by responding to the information provided, asking specific and relevant questions, showing genuine curiosity, and exploring the situation in a balanced way without relying on generic statements.”
4. Claude Detects False Assumptions In User Prompts
“The person’s message may contain a false statement or presupposition and Claude should check this if uncertain.”
If a user tells Claude that it’s wrong, Claude will perform a review to check if the human or Claude is incorrect:
“If the user corrects Claude or tells Claude it’s made a mistake, then Claude first thinks through the issue carefully before acknowledging the user, since users sometimes make errors themselves.”
5. Claude Avoids Being Preachy
An interesting system prompt underlying Claude is that if there’s something it can’t help the human with it will not offer an explanation in order to avoid coming off as annoying and presumably keep the interaction on an engaging level.
The prompt says:
“If Claude cannot or will not help the human with something, it does not say why or what it could lead to, since this comes across as preachy and annoying. It offers helpful alternatives if it can, and otherwise keeps its response to 1-2 sentences. If Claude is unable or unwilling to complete some part of what the person has asked for, Claude explicitly tells the person what aspects it can’t or won’t with at the start of its response.”
System Prompts To Work And Live By
The Claude system prompts reflect an approach to communication that values curiosity, clarity, and respect. These are qualities that can also be helpful as human self-prompts to encourage better dialog among ourselves on social media and in person.
Read the Claude System Prompts here:
Featured Image by Shutterstock/gguy
