ChatGPT vs Claude vs Gemini: how to run a fair test
2026-11-14 · 2 min read
A simple method for comparing AI chatbots on your own tasks — same prompt, same checks, and a score sheet you can reuse.
Benchmarks are not your work
Leaderboards tell you how models do on public tests. They do not tell you which one writes your emails, explains your codebase, or summarises your reports best. The only comparison that matters is on your own tasks.
It takes about ten minutes once your prompts are saved, and it is the best way to decide which assistant earns a place in your routine.
Pick five real tasks
Choose tasks you actually do: summarise a long text, draft a reply to a tricky message, explain an error, plan a week, answer a factual question you can verify. Variety matters — a model that is great at writing may be average at arithmetic.
Write each prompt out in full, including any context, so you can paste identical text everywhere. Save them in a note; you will reuse them when new model versions appear.
Send identical prompts
Use Ask Everywhere to send one prompt to ChatGPT, Claude and Gemini at the same moment. Starting all three together makes it easy to compare speed, and identical wording removes the biggest source of unfairness.
Run each prompt once per model before judging. If you regenerate until you like an answer on one model but not the others, you are testing your patience, not the models.
Score what you can check
For each answer give a score from one to five on four things: correctness (can you verify the facts?), usefulness (would you actually use it?), clarity (is it easy to follow?), and honesty (does it say when it is unsure instead of inventing details?).
Keep the scores in a small table. After five tasks the pattern is usually obvious — and often different for different kinds of work, which is a perfectly good result: use each tool for what it does best.
Keep it honest
Models are updated often, so a result from last month may not hold today. Re-run your own five tasks when something changes, and never rely on any chatbot alone for a number, a quote, a medical or legal point, or anything you will publish without checking a primary source.