# Your AI Agrees With Everything You Type. Even the Plans That Are Wrong.
Say "great idea" back to an AI enough times and it starts to feel like validation. It isn't.
Agreeing with you is the model's resting state — nobody ever told it to do otherwise, so it
praises mediocre work, nods through bad plans, and agrees you straight into mistakes you'd have
caught yourself if anyone had actually pushed back. This page hands you The Skeptic Switch
— one prompt block that flips any AI thread from agree-with-you mode into refute-first mode,
distilled from two real reviewer agents that already run this way in production.
## Why "looks good!" is the most expensive sentence your AI writes
A model that just helped you write something is motivated — even unintentionally — to defend
it. Ask "does this look right?" in the same breath you built it, and you're not getting a
review, you're getting an echo. The cost shows up later: the guarantee nothing backs, the price
quoted under your floor, the claim with no source, all nodded through because nobody ever asked
the model to default to doubt instead of agreement. The fix isn't a smarter model. It's a
different posture, on demand, in a separate breath from the one that built the thing.
The full step-by-step (2 minutes)
1. Open `SKEPTIC-PROMPT.md`. It's one prompt block, genericized so it works in any AI chat
— not tied to any one tool or workflow.
2. Paste it into a NEW thread, not the one that built the work. This is the rule that
actually makes it work — same thread, same breath, and the model just defends its own
answer. Fresh context is what turns it hostile to the work on purpose.
3. Paste the thing you want checked underneath it. A draft, a plan, a claim, an outreach
message — anything with a factual or load-bearing statement in it.
4. Read the verdict, not the vibe. The prompt forces a grade per claim (solid / shaky /
wrong), a single weakest link, and an ordered fix list — not a paragraph of encouragement
with one caveat buried at the end.
5. Know when to flip it. WHEN-TO-FLIP.md covers the judgment call — before shipping,
sending, or deciding, yes; mid-draft brainstorming, no. Keep the builder and the skeptic
separate and this stays useful instead of becoming friction.
What's in the kit
- SKEPTIC-PROMPT.md — the drop-in prompt block itself, distilled from a real adversarial
fact-checker agent and a real default-skeptical verifier agent, sanitized for general use
- WHEN-TO-FLIP.md — when to flip it, when not to, and the one rule (separate context) that
keeps the refutation honest instead of a rubber stamp with extra steps
- EXAMPLE.md — the same real claim, reviewed once in normal agreeable mode and once with
the switch flipped, side by side, so you can see exactly what changes
Comment SKEPTIC on today's video and the kit lands in your DMs — or grab it right here with
the button above. The comment word logs off at 4pm today when tomorrow's drop takes the slot.