Example: What Three AIs Told Me About Asking Better Questions
EXAMPLE POST. A short summary of asking ChatGPT, Grok and Gemini the same question: how do you write a prompt that gets a genuinely useful answer?
I asked the same question to three different models: “What makes a prompt actually good?” The answers overlapped more than I expected, and the differences were instructive.
Where they agreed
- Context beats cleverness. All three said the biggest lever is telling the model who you are, what you already know and what you’ll do with the answer.
- Show the shape of the output. Asking for “a table with columns X, Y, Z” or “five bullet points, no intro” gets more usable results than “explain”.
- Iterate in the open. Treat the first answer as a draft and push back on it.
Where they differed
ChatGPT leaned towards structure: roles, constraints, examples. Grok was blunter, suggesting I simply ask the model to argue against its own first answer. Gemini emphasised grounding: paste the source material rather than hoping the model remembers it.
The best prompt is often the second one, written after you’ve seen what the first one got wrong.
My distilled checklist
- State the goal and the audience in one sentence.
- Give the material: data, code, constraints.
- Specify the format of the answer.
- Ask for assumptions and uncertainties to be listed.
- Follow up with “what would make this wrong?”