When ChatGPT tells you Claude’s a better writer, you listen.

Ever since I started using Claude and ChatGPT, I've been fascinated by their differences. I use ChatGPT to organize my life, to build systems, to give advice, for prompt engineering -- my left-brain work.

But I always sensed Claude was the more sensitive writer. As one friend commented a few months back, "The ghost in Claude's machine feels more human."

On the day OpenAI released GPT-5, I was in the middle of crafting some content editor personae. In the morning, I was using GPT-4o — and it suggested creating separate prompts, optimized for the ways GPT and Claude “think.”

When I asked why, it told me what I’d long suspected: even ChatGPT believed Claude was a better writer. And it told me why it thought so.

But before I could fully process (and share) that insight, GPT5 dropped.

By the afternoon, GPT5 was in place — and it had news: it had closed the gap between itself and Claude. But there were still differences. Claude still has that mysterious sensitive site. And when I give both of them creative direction I’d offer a copywriter (i.e., surprise me with this headline. Zig when the reader expects a zag,”) Claude’s “zigs” feel more unexpected.

It’s useful to understand the nuances. So I asked ChatGPT to break it down:

Claude 3.5 vs. GPT-4 vs. GPT-5 — Writing & Editorial Strengths

• Sensitivity to tone/voice:

• Claude 3.5 — Strongest in emotional and narrative tuning.

• GPT-4 — Excellent but more literal.

• GPT-5 — Nearly matches Claude, with better tone preservation than GPT-4.

• Following nuanced role prompts:

• Claude 3.5 — Naturally flexible and conversational.

• GPT-4 — Strong when roles are clearly scoped.

• GPT-5 — Handles complex, multi-layered instructions with high accuracy while keeping tone intact.

• Length + reasoning tolerance:

• Claude 3.5 — Handles longform input smoothly with a free-flow feel.

• GPT-4 — May compress or summarize.

• GPT-5 — Manages longform well, keeps tighter structure than Claude.

• Editorial warmth:

• Claude 3.5 — Reads like a generous, intuitive peer.

• GPT-4 — Reads like a structured, practical coach.

• GPT-5 — Blends warmth with structured, precise feedback.

The bottom line — even LLMs have distinct quirks and superpowers, and each works better for different things. I’ll keep using both, liberally. But when it comes to copy editing, final draft, and smart headlines, I’m still “Team Claude.” And now I understand why.