Read the hype

“The world's smartest AI. Flawless. 10× faster.” Every launch sounds like magic. The skill isn't believing it or dismissing it — it's translating the pitch back into what the tool can actually do.

Spin has a grammar

Marketing language isn't random — it leans on a few repeatable moves. A superlative (“most powerful”) that names no rival. A benchmark score cherry-picked from the one test where the model wins. An absolute (“flawless”, “never wrong”) that no real system earns. A demo shown as if it were the everyday product. Spot the move and the claim shrinks to its true size.

A capability is not a demo

The gap that fools people most is demo versus product. A launch video is the best single run the company could capture, often hand-picked from many tries — your Tuesday-afternoon result is the average, not the highlight. “Can” is doing a lot of work too: “can write an app” means it managed to once, not that it reliably will for you. Read “can” as “sometimes, under good conditions” and you'll be right more often.

A three-question filter

Before you act on any AI claim, run it through three quick questions. Who is the comparison against, and on what test? Is this a controlled demo or the shipping product I'll actually use? And what would failure look like — has anyone shown the misses, not just the wins? A claim that survives all three is worth taking seriously. Most marketing quietly fails at least one.

• —Name the technique: superlative, cherry-picked benchmark, absolute, or demo-as-product.
• —Read “can” as “managed once, under good conditions” — not “reliably will”.
• —Ask who the comparison is against, and whether anyone showed the failures.