Sagan's Baloney Detection Kit, Updated for AI Slop

Carl Sagan wrote The Demon-Haunted World in 1995. Tucked inside it is a list called the Baloney Detection Kit: nine questions to ask before you accept a claim. Thirty years later it holds up better than 90% of the “media literacy” content written this decade.

I’ve been re-reading it because my feed is now 80% AI-generated productivity advice, hype-cycle screenshots, and growth gurus quoting each other. Sagan’s checklist is the cleanest filter I’ve found. Here are the nine, translated into 2026 versions, with the kind of slop they catch.

1. Independent confirmation of the facts

Sagan’s version: Wherever possible, get independent verification.

2026 translation: If a “study showed” is making the rounds and the only source is a screenshot of a Twitter thread, the study probably doesn’t exist or doesn’t say what the screenshot says. AI summaries hallucinate citations constantly. One source is rumor. Two unrelated sources is a tentative fact.

2. Substantive debate by knowledgeable proponents

Sagan’s version: Encourage debate on the evidence by all points of view.

2026 translation: If a productivity framework or AI tool has no serious critics (only ecstatic adopters and reflexive haters), you’re not looking at a real conversation. You’re looking at a marketing funnel. The good ideas have skeptics who can articulate the steelman. I came at this from a different angle in why most goal-setting frameworks are backwards.

3. Arguments from authority carry little weight

Sagan’s version: “Experts have committed errors in the past. They will do so again in the future.”

2026 translation: “Top performers do X.” “Founders all use Y.” “A senior engineer at a frontier lab said Z.” None of these are evidence. They’re appeals to credentials, which is how you sell a course, not how you check a claim.

4. Spin more than one hypothesis

Sagan’s version: Don’t simply run with the first idea that comes to mind. Think of all the ways it could be true and all the ways it could be wrong.

2026 translation: Before adopting an explanation (“my content isn’t growing because the algorithm hates me”), list three other explanations. Almost always, one of the other three is more boring, more accurate, and more actionable. The real reason your content isn’t growing is rarely the first one you reach for.

5. Try not to get overly attached to a hypothesis just because it’s yours

Sagan’s version: Ask why you like the idea. Compare it fairly with the alternatives.

2026 translation: If you’ve sunk three months into a side project and you’re suddenly very sure the market just needs more time, ask yourself whether you’d believe that if a stranger said it about their project. We over-defend our pet ideas because abandoning them hurts. Sunk cost masquerading as conviction is the most common form of self-baloney.

6. Quantify

Sagan’s version: If whatever it is you’re explaining has some measure, attach a number to it.

2026 translation: “AI saved me hours every week.” How many hours? Measured how? Compared to what baseline? Anyone telling you AI is transformative without producing numbers is selling. Anyone telling you it’s useless without producing numbers is also selling. Quantification is the cheapest BS detector we have.

7. If there’s a chain of argument, every link in the chain must work

Sagan’s version: Every link, not just most of them.

2026 translation: “If you build in public, you’ll get an audience, and if you have an audience, you’ll get customers, and if you have customers, you’ll have a business.” Four links. Each one fails for a lot of people. Long argument chains in productivity content collapse if you test any single link.

8. Occam’s Razor

Sagan’s version: When faced with two hypotheses that explain the data equally well, choose the simpler.

2026 translation: Your launch flopped. The complicated explanation is the platform’s algorithmic shift. The simple explanation is that the offer wasn’t compelling. Bet on the simple one until evidence forces you to upgrade. Most “the algorithm changed” theories die under this test.

9. Can it be falsified?

Sagan’s version: A claim that can’t be tested is worthless.

2026 translation: “Manifestation works if you really believe.” “The right opportunities will find you when you’re ready.” These claims have no losing condition. They’re not theories. They’re vibes. Any productivity advice that can’t be wrong isn’t actually advice. It’s reassurance.

How to actually use this

I keep these printed on the back of a sticky note next to my monitor. When I see a take that’s making me nod too hard, I run it against three or four of these questions before I share it, save it, or let it change my behavior. About a third of the time, the take collapses.

If you only remember one of the nine, make it the last one: can it be falsified? It does most of the work. If a claim can’t be wrong, it can’t be useful.

Sagan didn’t write this for the AI era. He didn’t have to. The kit was always meant for any era where attention is for sale and certainty is cheap. We just happen to be living in the most aggressive version of that environment so far. If you’re trying to build a habit of thinking clearly under pressure, starting with Sagan’s nine is the cheapest education available.