ChatGPT vs Claude: Which Should You Write Prompts For?

Both models reward the same fundamentals — context, structure, constraints — but they weight them differently, and prompts tuned for one often underperform on the other.

Where they differ in practice

Following constraints. Claude tends to treat constraint lists as hard rules; ChatGPT treats them more as strong suggestions. If exact word counts or banned phrases matter, Claude usually needs less repetition to comply.

Structure in the prompt. ChatGPT responds well to numbered steps and markdown headers inside the prompt. Claude handles both, and additionally responds well to XML-style delimiters (<context>, <task>) for separating sections in longer prompts.

Tone control. Both take tone instructions, but ChatGPT drifts back towards its default register in long outputs more often. Restating tone in the constraints section helps.

Refusals and hedging. Claude hedges more on uncertain facts, which is useful for research tasks and mildly annoying for creative ones. Adding “commit to a single recommendation” reduces hedging on both models.

The honest answer

For most everyday tasks the difference is smaller than the difference between a vague prompt and a structured one. Fix the prompt first; choose the model second.