Guides
Why OpenAI, Anthropic, Gemini, and Ollama Prompt Formats Differ
Two prompt templates can say almost the same thing and still need very different JSON shapes depending on the provider. That is because each API has its own way of modeling system instructions, message turns, and content parts.
Published March 22, 2026 · Updated March 22, 2026
Why The Shapes Are Different
Provider APIs evolved separately, so they do not all treat prompt structure the same way. OpenAI Messages, Anthropic Messages, Gemini contents, and Ollama chat payloads each organize conversation data a little differently.
That is why prompt migration often fails at the request-body level even when the core prompt text still makes sense.
Where Migrations Usually Break
Migrations often break around system instructions, nested content parts, or assumptions about role layout. A prompt that feels simple in ChatGPT-style JSON may need a different arrangement when moved into Claude-style or Gemini-style payloads.
Those differences matter most when you are trying to keep one prompt workflow reusable across several model backends.
Why Format Awareness Helps
Understanding the format differences makes prompt portability easier because you can review the converted structure instead of guessing how each provider wants the data arranged.
That is one reason provider-specific prompt converters are useful for real migration work.