The most important thing to understand about AI assistants is also the easiest to forget: they sound exactly as confident when they’re wrong as when they’re right.
An AI assistant works by predicting what words should come next, based on patterns in an enormous amount of text. When the patterns are strong, as they are for common, general questions, the answers are usually good. When the question is specific, recent, or about your situation, it can produce something that sounds perfectly reasonable and simply isn’t true. It doesn’t flag its own mistakes, and it doesn’t know when it’s guessing.
That doesn’t make these tools useless. It means you check. Here are five quick ways, in roughly the order I use them.
1. Ask it what to double-check
The simplest check is built in. After any answer that matters, ask:
What in your answer should I double-check, and where should I check it?
A good assistant will point you to the weak spots: the numbers, the dates, the parts that depend on your specific plan or state. It won’t catch everything, but it often points you straight at the claims most likely to be wrong.
2. Compare it with the original
If you asked the assistant to explain or summarize a letter, a policy, or an article, compare its answer with the original. Summaries can quietly drop the one exception that applies to you, or turn “may” into “will.” This takes a minute and catches a surprising number of problems.
3. Go to the official source yourself
For anything about your health, your money, or the law, check with the official source: the agency, the company, or the program. Type the website address in yourself, or use the phone number printed on your card, statement, or official letter.
Never use a phone number or website that an AI assistant gives you for anything involving your accounts or your money. Assistants sometimes produce numbers that look right and aren’t, and scammers are very happy to answer the wrong number.
4. Ask a real person who knows your situation
Your pharmacist, your doctor’s office, your bank, your accountant, a librarian, or a knowledgeable friend. AI is a good way to prepare for those conversations, to understand the terms and get your questions in order. It’s not a replacement for someone who knows your situation and is accountable for their answer.
In Everyday Help for the Rest of Us, a retired shop teacher named Frank nearly learned this the hard way. He asked an AI assistant whether a repair to his table saw was covered by the manufacturer’s warranty. It told him, clearly and kindly, that the brand offered five years on motors. He was halfway through an indignant email to customer service when his wife asked where the five years came from. The paperwork in his filing cabinet said three, and his had run out the previous spring. Now he says, “Trust, but verify the numbers.”
5. Watch for answers that are too neat
Be extra careful when an answer is exactly what you hoped to hear, includes very precise numbers or quotes with no source, or names a study, a law, or an expert you can’t find when you look. Invented details tend to look tidy. Real ones usually come with caveats.
How much checking is enough?
Match the checking to the stakes. A gift idea or a dinner recipe needs no checking at all. A summary of an article needs a quick comparison. Anything about health, money, legal matters, or a decision you can’t easily undo gets checked with an official source or a real person before you act.
That’s the middle step of the method in Everyday Help for the Rest of Us: Ask. Check. Decide. Use AI to ask and to understand. Check what matters. Then decide for yourself.
