Every new technology arrives wrapped in big promises. AI is no different, just louder. This site is a plain-language guide to spotting hype, what the evidence actually says, and how to write claims that hold up.
A claim is hype when it sounds measurable but isn't: no sample, no method, no date, no baseline. It's not always a lie. Often it's just a feeling dressed up as a fact.
The most useful test is simple. For any claim, ask whether you could check it today. If you can, it's proven. If you could with a test, it's testable. If there's nothing to check, it's hype.
The term reached regulators in 2024, when the US Securities and Exchange Commission brought its first "AI washing" cases.
Gartner's Hype Cycle describes five phases: a trigger, a peak of inflated expectations, a trough of disillusionment, a slope of enlightenment, and a plateau of productivity. It's a framework, not a law of physics, but it explains why the loudest claims arrive first.
In August 2026, Gartner summed up AI this way: "AI investment remains strong, but AI value realization is uneven." Money is still pouring in. Results vary a lot.
In a 2025 randomized trial, 16 experienced developers worked on 246 real tasks in code they knew well. Each task was randomly assigned to "AI allowed" or "AI not allowed."
Source: METR, "Measuring the impact of early-2025 AI on experienced open-source developer productivity" (July 10, 2025). METR describes it as a snapshot of early-2025 tools; results may differ for other people, tasks and newer models.
The lesson isn't "AI is useless." It's that self-reported results, the kind most vendor claims rely on, can be wrong in direction, not just size. When someone says their users "save 10 hours a week," ask how that was measured.
of organizations reported no business return from generative AI pilots.
MIT NANDA, The GenAI Divide (July 2025), as reported by Virtualization Review. Critics noted a small sample and a narrow definition of "return." Treat it as a signal, not a fact.
of generative AI projects forecast to be abandoned after proof of concept by the end of 2025.
Gartner (July 29, 2024). A forecast, citing poor data, risk controls, cost and unclear value.
of agentic AI projects forecast to be cancelled by the end of 2027.
Gartner (June 25, 2025). Also a forecast. Proven that Gartner said it; the outcome is testable until 2027.
Humane raised $230 million for its wearable AI Pin. Reviews called it slow and unreliable. The devices stopped working on Feb. 28, 2025, and HP bought the company's assets for $116 million.
The US Federal Trade Commission said DoNotPay couldn't back up that claim. The company agreed to pay $193,000 and stop making it, as part of the FTC's 2024 crackdown on deceptive AI claims.
Air Canada's website chatbot gave a customer wrong information about a bereavement fare. A Canadian tribunal ruled in 2024 that the airline was responsible for what its chatbot said, and had to pay.
And a twist: hype about hype. A viral 2025 story claimed Builder.ai's "AI" was really 700 engineers. Reporting later showed that story was itself overstated: the company did use AI tools, and its real problem was inflated revenue and insolvency (The Pragmatic Engineer, June 2025). Fact-check the takedowns too.
"Don't Panic" is printed on the cover of The Hitchhiker's Guide to the Galaxy. It's also the right attitude to AI hype: don't panic, check. Paste a claim, then answer the nine questions honestly.
Five claims. Five seconds each. Proven means you could check it today. Testable means a real test could check it. Hype means there's nothing to check.
The claims are examples written for this game, apart from the Gartner forecast.
You don't have to give up on bold claims. Make them checkable: say what it does, for whom, measured how, and when.
The numbers in the rewrites are illustrations. Use your own, with your own source.
Copy this into ChatGPT, Gemini or Claude, then paste the page you want to check. Read the output critically: AI tools can invent "real" numbers too.
GOAL: Check these product claims for hype. RETURN FORMAT: A table with: the claim (quoted), which DON'T PANIC questions it fails, a verdict (green, amber or red), and a one-line honest rewrite. WARNINGS: Judge only what the page shows. Don't guess the real numbers or invent sources. CONTEXT: [Paste the headline, the numbers and the testimonials from the page.]
The Competition Act bans false or misleading marketing, and courts look at the general impression a claim gives, not just the fine print. If a claim needs an asterisk, rewrite the claim.
The FTC's Operation AI Comply targeted deceptive AI claims and made the point plainly: there is no AI exemption from the laws on the books.
In March 2024 the SEC settled its first "AI washing" cases with two investment advisers, Delphia and Global Predictions, for misleading statements about their use of AI.
General information, not legal advice.
A promise you can't prove is hype.From the course The Hitchhiker's Guide to the AI Galaxy, Smith School of Business, Queen's University
A promise you can prove is a brand.