How to Move Claude Prompts to ChatGPT Message Format

Published March 18, 2026 · Updated March 18, 2026 · 3 min read

Moving prompts from Claude to ChatGPT is the reverse of the more common direction: you take the Anthropic top-level system parameter and move it back into the messages array as a system role message. The conversation content stays the same.

The structural conversion

Anthropic format with a system parameter:

{
  "model": "claude-3-5-sonnet-20241022",
  "system": "You are a helpful assistant that summarises documents concisely.",
  "messages": [
    {"role": "user", "content": "Summarise this: ..."},
    {"role": "assistant", "content": "Here is the summary: ..."},
    {"role": "user", "content": "Make it shorter."}
  ]
}

OpenAI format with the same content:

{
  "model": "gpt-4o",
  "messages": [
    {"role": "system", "content": "You are a helpful assistant that summarises documents concisely."},
    {"role": "user", "content": "Summarise this: ..."},
    {"role": "assistant", "content": "Here is the summary: ..."},
    {"role": "user", "content": "Make it shorter."}
  ]
}

The system content moves from a top-level field into the messages array as the first message with "role": "system". The conversation messages are identical — only the container changes.

What doesn't change

User and assistant message content transfers directly. Both APIs use the same user and assistant role names. Content as a plain string works in both. Multi-turn conversation structure is the same in both formats.

What to adjust beyond the structure

The model name changes (from a Claude model to a GPT model). The max_tokens parameter in Anthropic becomes max_tokens in OpenAI — both use the same parameter name. Anthropic's temperature and top_p work the same way in OpenAI.

Tool/function calling syntax differs significantly between providers — if your prompt uses tools, this requires separate handling beyond the message format conversion.

Output differences are normal: GPT-4o and Claude 3.5 Sonnet are different models with different characteristics. A prompt converted from Claude to GPT-4o will produce equivalent but not identical output. Test with representative inputs and adjust system instructions if the output style differs in ways that matter for your use case.