A practical comparison of Perplexity and ChatGPT for research tasks — not benchmarks, but where each one actually holds up and where it quietly fails.