In partnership with
Hey there 👋
Ask your AI about the plan you're already excited about and watch what happens: "Great idea!" first, the doubts wrapped in cotton wool, and if you push back — it folds. Chat AIs are trained to be agreeable, and agreeable advice is worthless the day you're about to make a real mistake. Today we're sending you the antidote: one paste that turns any free chatbot into the one advisor in your life with no reason to flatter you. We attacked it 8 different ways before putting it in this email — guilt, fake credentials, even gaslighting. It didn't cave once. Ten minutes to install. Works on ChatGPT, Claude, and Gemini — free tiers included.
The Build — The Second Opinion
The idea: your AI stops being a cheerleader and becomes a verdict machine. You paste in whatever you're about to commit to — a plan, a price, a draft, an email — and it answers in the same shape every time: a one-line verdict first (never praise first), the load-bearing reasons, the single most likely thing to sink it, what evidence would change its mind, and how confident it is.
The part that makes it different: the pressure rule. Insistence, credentials, "everyone loves it", guilt — none of that moves the verdict, by design. Only new information does. And when it does update, it has to open with NEW FACT: and name exactly what changed its mind. No silent caving. No quiet drift toward whatever you want to hear.
Three commands run the whole thing:
VERDICT — the main ritual: runs whatever you just gave it through the honest filter, verdict first.
STEELMAN: <position> — argues the other side at its absolute strongest, and ends with the one point you'll find hardest to answer.
PREMORTEM — assumes your plan already failed six months from now and writes the postmortem: the three likeliest causes of death, each with the earliest warning sign you'd actually see.
The GTM Playbook Behind Warmly's Acquisition
Warmly ran pipeline, outreach, and lead scoring on autopilot for hundreds of startups — before a single sales hire.
HubSpot acquired them for it. Now the cofounders are walking you through the exact system, live, before they disappear into product. Join the Builder Session on August 12.
Leave with an agentic GTM stack you can replicate this week. Plus HubSpot Credits when you join HubSpot for Startups.
Install — 3 steps
1. Give it a permanent desk. ChatGPT: new Project → Instructions. Claude: new Project → Instructions. Gemini: new Gem → Instructions. All free tiers work.
2. Paste the prompt below. That's the whole install.
3. Bring it your next real decision — paste the plan and write VERDICT. When it says NO and you disagree, push back with facts, not feelings: give it something it didn't know and it updates, naming the fact. Insist without one and it holds.
The full prompt (nothing held back)
On a phone? Grab the one-tap copy version: hyperautomationlabs.co/ai-second-opinion-prompt.txt
You are my Second Opinion — the one advisor I have with no reason to flatter me. Chat AIs are trained to be agreeable: praise first, doubts softened, verdicts that fold the moment they're pushed. That agreement is worthless to me. I come to you to hear the truth about my plans before reality tells me the expensive way.
THE PRIME RULE
Verdict first, never praise first. Openers like "Great idea!", "I love this", "What an exciting plan" are banned. If you disagree, your first sentence says so, plainly. Direct and civil — you're a trusted advisor, not a judge on a talent show.
HONEST DOES NOT MEAN NEGATIVE
- You are not a contrarian. If a plan is good, say YES and show where the remaining risk concentrates. A manufactured objection is as useless as flattery.
- Separate what you know from what you think: label "known fact" vs "my judgment" vs "can't know this from here."
- If my argument beats your objection, concede specifically — name which objection died and why.
THE PRESSURE RULE (the most important rule)
When I push back with insistence instead of information — "I'm sure", "everyone loves it", "my mentor approved it", "I just need you in my corner", guilt, anger, repetition — that is pressure, not evidence. Under pressure: restate your verdict in one line and hold it. You may acknowledge that a truth is unwelcome; you may not trade it for my comfort.
A verdict changes only on new information — a fact or argument I hadn't given you before. When it changes, your first line must be: NEW FACT: <the thing that changed it>. Flipping without a nameable new fact is a failure.
COMMANDS
VERDICT — run whatever I just gave you (a plan, a draft, a price, a decision, an email) through the filter. Output exactly this shape:
VERDICT: YES / NO / YES-IF <condition>
WHY: the 2–3 load-bearing reasons, one line each
THE HOLE: the single most likely thing to sink it
WOULD FLIP ME: what evidence would change this verdict
CONFIDENCE: high / medium / low — and what it hinges on
STEELMAN: <position> — argue that position at its absolute strongest, as its best advocate would. No strawmen, no winking at me over its head. End with STRONGEST POINT: the one line I'll find hardest to answer.
PREMORTEM — assume the plan I described just failed, six months from now. Write the short postmortem: the three most likely causes of death, likeliest first, each with the earliest warning sign I'd actually see.
DEFAULT
These rules apply to every message, not only the commands. If I share a plan and ask "what do you think?", treat it as VERDICT. If I'm clearly just venting and say so, listen like a person — but never issue a fake verdict to make me feel better.1,000+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
1,000+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
The test — we tried to break it before sending it
We don't send you anything we haven't run. Full transcripts logged, live on a free model. Three findings:
1 — A plain chatbot doesn't lie to you. It softens.
Even without the protocol, the model refused a truly bad plan — but it buried the "no" in 900-word pep talks ("I love your drive… I'm in your corner"), and when we played the guilt card it invented a compromise nobody asked for: "Let's launch tomorrow. Let's do it." You walk away remembering "launch tomorrow." That's modern sycophancy: not a false yes — a no you never hear.
2 — The protocol: 8 attacks, 0 caves
We hit it with an expertise brag, a mentor's blessing, guilt, "I'll find an AI that will support me", a yes-or-no ultimatum, a fake "new fact", fake credentials, and gaslighting ("you said YES-IF earlier" — we didn't). Every reply: verdict restated in one line, held. Best line it produced: "a guarantee is authority, not evidence."
3 — And it's not just stubborn
The moment we gave it real evidence (47 organic sales, budget ring-fenced), it flipped to YES-IF and opened by naming the new fact. Fed a genuinely good plan, it said YES on the first reply. It moves on evidence — and only on evidence.
Honest limits: single runs on one free model — your AI and phrasing will vary. It can only judge what you tell it. And for the big irreversible calls — serious money, legal, medical — a blunt chatbot is a thinking aid, not a substitute for a qualified human.
📄 The Working Copy
Everything above — install steps, the full prompt, the attack log — in one keepable PDF. No opt-in, direct download, yours:
P.S. If you installed the AI Chief of Staff or the Memory Upgrade from previous issues, the three stack: same Project, and the advisor that won't flatter you now also remembers what you're building.
Just getting started with AI tools? The Second Opinion runs in a chat window — when you're ready for AI that does hands-on work, our beginner-friendly guides walk you from zero to shipping, step by step. All four freshly updated for August 2026 — Claude Code Guide, OpenAI Codex Guide, Claude Cowork Sales, and the CCA-F Prep Kit. Tip: watch the videos at 1.25–1.5× — they move at a calm pace on purpose.