Can you trust AI? How to check an answer

It’s probably the question people ask most often: can you trust AI? The answer fits in one sentence. You can, as long as you don’t trust it blindly. An AI is like a very knowledgeable friend who is always available, has an answer for everything and sometimes talks with great confidence about things they barely know. This article helps you see what needs checking and suggests a method to adapt to how important your decision is.

It depends on what you do with the answer

AI produces the answer that seems most likely to it. It phrases it with the same confidence whether it’s right or wrong, so the tone tells you nothing about reliability. To understand why it sometimes makes things up, read our article on AI hallucinations.

Before asking whether AI is reliable, ask yourself what you’re going to do with its answer. The more your decision commits you, the more checking matters.

What to check, depending on the use

• When AI rephrases, summarises or translates a text you give it, check against your original that nothing has been added or left out, especially figures, dates and names.
• For ideas for a meal, an outing or a gift, match your checks to your constraints and to what a mistake could cost. Just check what commits you, such as a time or a price, on the venue’s or seller’s website.
• When it explains a subject, check the facts you’re going to rely on against a reference source, such as an organisation’s website or a recognised encyclopaedia.
• Figures, dates, quotes and references all need checking, at the original source rather than on a site that repeats it.
• For health, money, legal matters or official procedures, the answer is a starting point. The decision is made with the official organisation or a professional.
• For news and recent topics, look at the date of the information first, then check with a recognised media outlet or the official website concerned.

The 5-step method, on a worked example

Imagine you’re planning a day out and ask the AI for the opening times of a museum near you. It replies “The museum is open every day from 10 am to 6 pm, entry £12”.

1. Spot what will guide your decision, here “every day”, “10 am to 6 pm” and “£12”. The museum’s history, which it tells you next, matters much less.
2. Ask where the information comes from, for example “What’s your source for these opening times?”. If it cites a website, open it. If it cites none, the information still needs checking.
3. Go to the museum’s official website. You might find that it’s closed on Mondays and that the price has changed, two details that change often.
4. Ask it what might have changed. It usually flags bank holidays, summer opening hours or compulsory booking on its own.
5. Decide with the right information. Keep the idea and the plan AI gave you, and correct what was wrong.

In practice, it takes two minutes, and only for what matters.

Questions to ask the AI to check

Keep these four questions handy. They can help you spot points to check, without guaranteeing that all of them will be flagged.

• “What are your sources? Give me the links.”
• “What in your answer could be wrong or out of date?”
• “What are the arguments against?”
• “If you’re not sure, say so.”

Always open the links it gives, as an AI can also cite a source that doesn’t exist, or make it say something else.

We don’t all trust in the same way

With information, some people cross-check without thinking about it. They’re the verifiers, like the Investigator or the Guardian. Others trust more readily and save their energy for action, like the Pioneer or the Pillar. Checking and moving forward are two complementary needs. Everyone can adjust their practices to the context, whatever their profile.

Knowing where you stand helps you check at the right moment, without overdoing it. It’s one of the three dimensions measured by our AI profile test. In about 3 minutes, you’ll know whether you lean towards verifier or trusting, and you’ll get a tip suited to you for making better decisions with AI.