How do I read an AI headline?

Two headlines in one morning's feed. The first says an artificial intelligence (AI) model just passed a famous exam. The second says the same kind of model can't count the letters in a word. It's tempting to decide one of them is wrong. Often both are true, and the trick is in what each one leaves out.

AI news swings between amazing and alarming, because both get clicks. A headline has a few words to make you stop scrolling, so it reaches for the most striking version of the story. The details that would calm it down, like how the test was run or what the system still gets wrong, end up far down the page, or out of the story altogether. You don't need to be an expert to put them back. A few plain questions do most of the work.

Here they are, short enough to keep in your head.

• Who is saying so?
• What exactly was measured?
• Was it a demo, or something you can use?
• What does the original source say?

Take them one at a time.

Start with the source. A lab announcing its own model, a company selling a product, an independent researcher, and a critic writing an opinion piece can all be honest, and each has a point of view. A lab's announcement tends to show its best results. A critic may lead with the worst ones. Neither is a reason to dismiss the claim. It's a reason to look for someone who has checked it and has nothing riding on the answer.

"Passed a famous exam" usually means one model scored well on one test, under conditions the testers chose. Was it the real exam, or practice questions? Did it get one try, or many? Did it have the same time limit people do? A result can be true and still say little about everyday work. Topic 67 explains how AI tests are built, and why they wear out.

A demo is a show. It's planned, rehearsed, and often run on the examples that work best. A product is what you can open on your own phone today and try on your own messy requests. The distance between the two can be large, and a demo can take months or years to become a product, if it ever does. So when a headline says AI "can now" do something, look for whether anyone outside the company can use it yet.

When a story mentions a study, it often links to it, and the study usually claims less than the story does. Researchers tend to write carefully: in our tests, on this task, more work is needed. Those limits can get lost on the way to a headline. You don't have to read the whole paper. The summary at the top and the section on limitations take a couple of minutes, and they often settle the question.³

If AI news leaves you unsure how to feel, you have company. In surveys reported in Stanford's AI Index 2026, about half of adults across some thirty countries said AI products made them nervous, and about half said they were excited by them. Plenty of people feel both. The questions don't pick a side for you. They show you what's there. Try them on the next AI headline in your feed, and notice which question it can't answer.¹²