← The AI Hype Audit — all 296 verdicts
REAL Real: the '9 prompts to stop AI making the decision for you' are sound decision hygiene, with one prompt that asks the model for a number it cannot honestly give
The claimEleven-slide carousel, '9 prompts to stop AI making the decision for you. It's very good at sounding certain. The expensive part was never the wrong answer; it's that I stopped comparing the second it picked one.' The nine: Give me the trade-off ('Don't recommend one. List the three options and what each one costs me'); Show your working ('What exactly are you basing that on, and what would change it?'); Argue the other side ('Now make the strongest case against what you just told me to do'); Name what you assumed ('List every assumption you made about my situation that I never told you'); Tell me what breaks ('If this is wrong: what breaks, when do I find out, and can I undo it?'); Say what most people do ('What's the boring default here, and what would have to be true to beat it?'); Rate your certainty ('Score your confidence 1 to 10, and say what's pulling the number down'); Ask what I left out ('What's the question I should be asking that I haven't asked?'); Make me say it ('Don't answer yet. Ask me the three questions you need answered first'). Final slide: 'Save this for later and follow for daily AI prompts.'
This is one of the better prompt carousels to pass through the audit. The premise is correct: models are fluent and confident by default, and the failure mode the poster names, accepting the first recommendation and ceasing to compare, is the one that costs real money. Eight of the nine prompts are standard, well-founded techniques: forcing options with costs, asking for evidence and what would change it, steelmanning the opposite, surfacing hidden assumptions, pre-mortem and reversibility, base rates, the missing question, and having the model interview you before answering. They work in any capable model and the poster sells nothing. The one that needs a caveat is 'rate your certainty 1 to 10': a model's self-reported confidence is not a calibrated probability, and the number it gives tends to cluster high regardless of accuracy. The useful half of that prompt is the second clause, 'say what's pulling the number down', which produces the actual caveats; the number itself should be treated as decoration.
What holds up
- All eleven slides read via logged-in Chrome (?image_index=N navigation); the nine prompts above are transcribed from the slides.
- Eight of the nine map directly to documented reasoning techniques (option enumeration, evidence and defeaters, steelmanning, assumption surfacing, pre-mortem and reversibility, base-rate defaults, question generation, clarify-before-answer). None depends on a specific model or product.
- Verbalised confidence scores from language models are known to be poorly calibrated and skewed high; the 'score 1 to 10' prompt yields a number that should not be read as a probability.
- Caption pulled via TikTok oEmbed; the only ask is 'save this' and a follow. No product, course or comment keyword.
What doesn’t
- 'Score your confidence 1 to 10' produces a number the model cannot honestly give; keep the 'what is pulling it down' half and ignore the digit.
- None otherwise. The post sells nothing and claims nothing it does not deliver.
The catch
Use these. They are the working habits of people who get good output from models, written as nine one-liners. Skip the confidence score and keep the question behind it.
How to actually do it
- Put the three you will actually use into your saved preferences or a text shortcut: 'list three options with costs', 'list every assumption you made about me', and 'ask me three questions before answering'.
- For any decision with money attached, run 'tell me what breaks and can I undo it' before acting. Reversibility is the cheapest risk control there is.
- Replace 'rate your confidence' with 'what evidence would change this answer'; you get the caveats without a fake number.
- Do the comparing yourself. The post's own insight is the real one: the model's job is to widen the options, yours is to choose.
Nine sound, model-agnostic prompts for keeping the decision with the human, sold with nothing attached. One prompt asks for a confidence score the model cannot calibrate; keep its second half.
- Confidence
- High
- Posted by
- @dj.bomasterflex ('AI Tools Daily'; photo post sent to the self-thread Thu Sep 10, 2026, 10:3x PM)
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.