How to spot AI hype (and why most of it is hype)

We read a lot of AI coverage so you don’t have to. Here’s our shortlist for spotting the stuff worth your time and ignoring the rest.

Tell #1: The piece doesn’t name what AI is doing differently

Watch for words like “AI is transforming everything” or “the AI revolution is reshaping work.” Then ask: what specifically? Which task? Which workflow? What’s the before, what’s the after?

If the piece can’t name a specific, mechanical difference, it’s probably not reporting. It’s probably commentary.

Tell #2: The numbers are vague or impossible

“AI is going to add $X trillion to the economy by 2030.” OK — based on what model? Adoption rates from where? Compared to what counterfactual?

There are real economic forecasts being done. They have caveats wider than highways. If a piece quotes a single round number with confidence, the confidence is the problem.

Tell #3: The author has something to sell

This isn’t disqualifying — lots of useful coverage comes from people building in the space. But it’s worth knowing. A McKinsey report on AI productivity is going to find AI productivity. A consultant who sells “AI transformation services” is going to find that you need AI transformation services.

Read the byline. Read the affiliations. Adjust accordingly.

Tell #4: The piece never mentions anyone who’s tried and failed

Real reporting includes failure. Hype writing doesn’t.

If every example in the piece is a glowing success, every executive is “excited about the opportunity,” and every worker is “embracing the new tools” — that’s marketing copy with a journalism layer. Real workplaces have screw-ups, half-baked rollouts, and people who quit because the AI made their job worse, not better. Coverage that omits that isn’t honest.

Tell #5: There’s no caveat

This is the simplest and the most reliable. Every honest AI take we’ve read in the last year contains the phrase “we don’t yet know” or “this is changing fast” or “there are real limits” somewhere in it. Hype takes don’t have those phrases.

If a piece is 100% confident, that’s the piece to skip.

What to read instead

We try to point at the takes that pass these tests, when we find them. The site’s feed page features the ones we’ve published. The newsletter sends one when something genuinely useful crosses our desk.

We’d rather under-promise. There isn’t actually that much great writing in this space yet. But the share is rising, and the signal is finding its readership.