I Used ChatGPT, Claude, and Gemini to Build the Same Project. Here's What Actually Happened

Same prompt. Three AIs. Wildly different code.

I gave all three the same task: build a REST API with auth, rate limiting, and a simple dashboard. No extra instructions. No hand-holding.

ChatGPT dropped a working Express setup immediately:

const rateLimit = require("express-rate-limit");
const limiter = rateLimit({ windowMs: 15 * 60 * 1000, max: 100 });
app.use("/api/", limiter);

Clean. Opinionated. Ready to run.

Claude went the other direction — asked two clarifying questions before writing a single line. Annoying at first. But when the code came, it was structured like a real codebase, not a Stack Overflow snippet.

Gemini gave me this:

from flask_limiter import Limiter
from flask_limiter.util import get_remote_address

limiter = Limiter(app, key_func=get_remote_address, default_limits=["100 per 15 minutes"])

Wait — Python? I never said Python. It just... decided.

Three models. Three assumptions. Three completely different starting points.
Then I pushed further. Same follow-up prompt for all three:

"Add JWT auth and protect the /dashboard route."

ChatGPT patched onto its existing code. Claude refactored the whole structure. Gemini switched frameworks mid-conversation without warning.

One of them caused a subtle auth bug I almost shipped.

The differences didn't stop at code style. How they handled edge cases, incomplete prompts, and outright wrong assumptions...that's where things got genuinely interesting. And occasionally, genuinely scary.

I documented every prompt, every response, every failure.
The full breakdown, including the auth bug, which model I'd actually trust in production, and the one prompt that broke all three...is in the blog post.

🤖 Explore it: View project and blog

If you've run your own AI head-to-head, drop it in the comments. Curious if others hit the same Gemini quirk.