CyberSandberg

AI

How to ask AI a good question, and get an answer that actually helps

Checked

Ask a chatbot "What should we make for dinner?" and you get a list that could be for anyone. Ask "Suggest three weeknight dinners for a family of four with a picky 8-year-old. Each should take under 30 minutes and have no peanuts. Finish with one shopping list," and you're more likely to get something you can cook tonight. Same tool. The difference is the question.

Talk about it at home: jump to the Kitchen Table Talk

You don't need special words for this. OpenAI's own prompting guide says "You don't need technical syntax or a rigid formula." What helps is the same thing that helps when you ask a person: say what you need, who it's for and how you'll use it, then keep talking.

The examples below come from the help pages of OpenAI (ChatGPT), Google (Gemini), Microsoft (Copilot) and Anthropic (Claude). They are examples, not a ranking or an endorsement. The same habits work with any AI tool.

1. Give context, but not private details

When an answer is too generic, Anthropic's guide says to "Add details about your audience, role, or constraints." Google's guide for Gemini in its Workspace apps says to "Provide as much relevant background as possible."

Before: "Is this a good laptop?"

After: "I'm choosing a laptop for a high school student who mostly writes papers and joins video classes. Our budget is about $500. What should I look for?"

Notice what the better question leaves out: names, the school, account numbers or anything you wouldn't say to a stranger. Context is about the situation, not your identity. For what's safe to share, see What to tell a chatbot, and what to keep to yourself.

Do this now: Before you hit send, add one sentence about your situation and check that it has no private details.

2. Say who it's for and what you want back

OpenAI suggests including "the audience or format when those details change what ChatGPT should produce." Microsoft's Copilot guide describes four parts of a prompt: a goal, context, a source and expectations ("How should Copilot respond?"). It adds that "all that's required is a clear goal," but you will often need more to get what you want.

Before: "Explain inflation."

After: "Explain inflation to a 12-year-old in three short sentences, with one example using a grocery item."

Format matters as much as audience. A table, a checklist, a short email under 120 words or a day-by-day plan are all fair requests.

Do this now: End your next question with the format you want, such as "in a short list" or "in a table."

3. Ask for options and tradeoffs

A single answer can hide the choices behind it. OpenAI's guide gives this example: "Compare these two phone plans for one person who travels internationally twice a year. Show the important differences in a table, then recommend one and explain the tradeoff."

Before: "What's the best way to save for a car?"

After: "Give me three ways to save about $3,000 for a used car in 18 months. For each, list one upside and one downside in plain language."

Asking for options keeps you in charge of the decision. You see what you would give up with each choice.

Do this now: Add "Give me two or three options, with the upside and downside of each."

4. Ask for sources, and ask it to say when it's unsure

Chatbots can sound sure and still be wrong. Microsoft says "Occasionally, Copilot responses can include incorrect content." NIST, part of the U.S. Department of Commerce, warns that AI tools can "confidently present erroneous or false content," including "citations" that only seem to support an answer.

You can ask the tool to help you check. OpenAI suggests asking "for sources when you need to check the result," and, for important work, asking it to flag "information it couldn't verify." Anthropic's guide suggests you "Ask Claude to cite sources or indicate confidence level."

Before: "How much water should my dog drink a day?"

After: "About how much water does a healthy adult dog need each day? Explain your reasoning, link the sources you used, and tell me if any part is uncertain or should be checked with a vet."

A source list, like the chatbot's own confidence, is a starting point, not proof. Open the links and make sure they exist and say what the chatbot claims.

Do this now: Add "Show your sources, and tell me what you're unsure about."

5. Follow up and push back

The first answer is rarely the last word. OpenAI says "Your first prompt doesn't need to be perfect." Google's guide says "use follow-up prompts and an iterative process of review and refinement," and Microsoft says to "Expect some back-and-forth conversation."

Good follow-ups are specific. Anthropic notes that "Make it shorter" is fine but a precise instruction is better. You can also disagree:

If the chat goes off track, Anthropic says it's sometimes faster to start a new chat with a clearer question.

Do this now: After the next answer you get, ask one follow-up that corrects, narrows or questions it.

Then check the answer

A better question gets a better answer, but not a guaranteed one. Microsoft notes that the same prompt can give "slightly different results each time," and says to "cross-reference with trusted sources when needed." For anything about health, money, law or safety, confirm the key facts with a source you trust or a qualified professional. Our guide to checking an AI answer walks through how.

Our rule of thumb: ask the way you would ask a smart helper who has never met you, and check the answer the way you would check advice from a stranger.

Teach it this week

AI

How do you ask AI a good question?

Big idea: Vague questions get vague answers. Add context, ask for options and sources, then check.

Background for the teacher or parent: Microsoft says a prompt can include a goal, context, a source and expectations. OpenAI suggests asking "for sources when you need to check the result." See Microsoft: Get started writing prompts and OpenAI: Prompting.

Classroom version (10 minutes)

  • Objective: Turn a vague question into a clear one, without private details.
  • Hook (2 min): Compare "Tell me about volcanoes" with "Explain how volcanoes form to a fifth grader in three sentences." Which helps more?
  • Explain (3 min): Four upgrades: who it's for, what format, options, sources. No names or personal details.
  • Activity (4 min): Pairs upgrade three vague questions on paper, two upgrades each.
  • Exit ticket (1 min): Write one follow-up to ask if an answer seemed wrong.

Kitchen Table Talk

  • When has someone misunderstood your question? What was missing?
  • How is a question for a friend different from one for a stranger?
  • Why might a computer sound sure even when it's wrong?
  • How would you check an answer you weren't sure about?

Try it together (no screens needed): Play "Make the question better." Someone says a vague question, like "What's a good book?" Each person adds one detail (age, length, topic) until it's clear. No real names.

Family action: This week, ask one real question the better way together, then check one fact. Our rule of thumb: if a helper who's never met you couldn't answer it, add more context.

Use made-up details, never type passwords or personal information into AI, and check each tool's age rules. Free to use; please keep the source credit.

This lesson also has its own page: How do you ask AI a good question?.

Education only, not financial, legal or tax advice. Not a registered investment adviser.

Sources

Primary sources

  1. OpenAI: Prompting (ChatGPT guide), date checked: Oct 22, 2026
  2. Google Docs Editors Help: Learn about prompts for Gemini in Docs, Sheets, Slides, Vids & Forms, date checked: Oct 22, 2026
  3. Microsoft Support: Get started writing prompts in Microsoft Copilot, page says "Last updated: September 2026," date checked: Oct 22, 2026

Supporting links

  1. Anthropic Claude Academy: Claude 101, Getting better results, date checked: Oct 22, 2026
  2. NIST AI 600-1: Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, July 2024, date checked: Oct 22, 2026