The Four Questions That Beat Any Fake Stat
Here's an uncomfortable truth: most of what gets called "news" or "research" today is somebody trying to sell you something. Not always a product. Sometimes an idea, a program, a version of the future they want you to buy into. Doesn't matter — the tell is the same.
You don't need a journalism degree or a PhD to catch it. You need four questions.
THE FOUR QUESTIONS
Who paid for what your seeing?
How many people or cases did they actually look at, and how did they look at them?
When did this come out?
Has anyone else, with no connection to the first group, checked it and found the same thing?
That's the whole test. If a claim can't answer these, it's not information. It's an opinion wearing a lab coat.
CASE ONE: THE "80% OF COMPANIES ARE FAILING" NUMBER
You've probably heard some version of this: most companies that try to use AI are failing at it. It gets repeated everywhere — LinkedIn posts, sales pitches, podcasts telling you their program is the fix.
Run it through the four questions.
CLAIM: Most AI projects fail.
VERDICT: VERIFIED
FINDING: A real research group — RAND Corporation, which has no product to sell you — studied this in 2024 and found that more than 80% of AI projects never make it to real use. MIT ran a separate study in 2025 with its own team and found something similar: about 95% of AI trial programs showed no real financial payoff.
RECOMMENDATION: You can trust this number. Two separate groups, different years, different methods, same conclusion. That's what verification looks like.
[MIDDLE LINE] Here's the part almost nobody tells you: the number is real, but the story built on top of it usually isn't.
CLAIM: Companies fail at AI because old-fashioned management and hierarchy are now obsolete, and the fix is a total company redesign.
VERDICT: HYPE
FINDING: RAND's actual explanation is boring — companies fail because of bad leadership follow-through and messy data, not because "hierarchy is dead." Nobody has independently confirmed the bigger claim. It's one person's forecast, dressed up next to a real statistic so it feels just as solid.
RECOMMENDATION: Don't repeat it as fact. A true number sitting next to a guess doesn't make the guess true.
This is the most common trick in modern content: take one number that's actually been checked, and staple an unchecked, much bigger claim right next to it. Readers assume both are equally solid. They're not.
CASE TWO: WHY "PEER-REVIEWED" DOESN'T MEAN WHAT YOU THINK
For decades, "it's peer-reviewed" was the gold standard. A study got checked by other experts before it was published, so it must be solid.
Here's the problem: in 2016, a large survey of scientists found that most of them had tried and failed to repeat another scientist's experiment and get the same result. Whole fields — nutrition science, psychology, some areas of medicine — have gone through periods where a large share of famous, peer-reviewed findings didn't hold up when someone else tried to repeat them.
CLAIM: If a study is peer-reviewed, the finding is true.
VERDICT: FALSE
FINDING: Peer review checks whether a paper follows the rules and makes logical sense. It does not check whether the result is actually true, and it doesn't require anyone to repeat the experiment. Reviewers are often a small, tight circle of people in the same field, sometimes with their own funding or reputation tied to similar work.
RECOMMENDATION: Treat "peer-reviewed" as step one, not the finish line. Ask the four questions anyway.
This isn't an argument for trusting nothing. It's an argument for trusting the process, not the label.
WHAT ACTUALLY HOLDS UP
Not everything is spin. Some sources are close to as objective as it gets, because nobody involved is trying to sell you anything:
Government data — things like the Census Bureau, the Bureau of Labor Statistics, the Federal Reserve's small business surveys. Nobody's getting a commission off these numbers.
Court records and government filings — facts stated under legal penalty.
Independent, nonprofit research groups with no product line, like RAND.
A study that's been repeated by a second, unconnected team and got the same answer.
Everything else — vendor research, trade publication "studies," most of what shows up in your inbox with a stat in the subject line — should be treated as a pitch until it clears the four questions.
THE TAKEAWAY
You don't need to become a skeptic about everything. You need one habit: before you believe a number or repeat it to a client, a friend, or your own team, ask who funded it, how they checked it, when it came out, and whether anyone unconnected got the same result. If you can't answer two of those four, don't repeat the claim as fact. Call it what it is — unverified, or hype — and move on.
The four questions don't make you paranoid. They make you the only person in the room who actually checked.