Before you trust an AI answer: a five step check

AI tools are very good at producing answers that sound right. That is a different thing from being right. They can invent a source, misremember a number, blend two facts into one, or describe a study that does not exist, all in the same calm tone they use when they are correct. Confidence is not accuracy.

You do not need to distrust everything. You need a habit that matches the effort to the stakes. This check takes about two minutes and works for a homework answer, a health question, a news claim, or a message you are about to send.

1. Ask what kind of question this is

Some questions have checkable answers (a date, a definition, how something works). Some are opinions dressed as facts. Some change over time (prices, who holds a job, what a law says right now). AI tools are weakest on anything recent, specific, numeric, or niche. If your question is one of those, you already know to check harder.

2. Ask for the source, then go look

Ask the tool where the claim comes from. Then open the source yourself. Two things can go wrong: the source does not exist, or it exists and does not say what the tool said it says. Either one means the answer fails. A source you did not open is not a source. It is a decoration.

3. Cross check with one independent source

Independent means not another AI tool and not a page that copied the first one. A library database, an official site, a primary document, a person who knows. One good independent confirmation is worth more than five identical search results.

4. Look for what is missing

Ask the tool directly: what would make this answer wrong? What am I not asking? A good answer to a bad question is still a bad answer. This step catches the cases where the tool answered the question you typed instead of the question you meant.

5. Decide the stakes, then decide what to do

• Low stakes (a brainstorm, a first draft, a summary you will read yourself): use it and move on.
• Medium stakes (schoolwork you will turn in, something you will tell a friend): do steps 2 and 3.
• High stakes (health, money, legal questions, anything you will publish or send to someone who will act on it): verify every claim you rely on, or do not act on it.

The habit underneath

The point is not to catch the AI. The point is to stay the one making the decision. People who use these tools well are not the ones who trust them least. They are the ones who know exactly when to slow down.

You checked before you shared. That is the whole skill.

For families: pick one real question a week and run the check together. Let the kids find the source. It takes ten minutes and it transfers.