Ask AI for a second opinion that pushes back.

Timeabout 10 minutes to read, 2 to try one prompt

Cost$0 extra, but you need an AI chat (Claude Code needs a paid Claude plan)

When you ask an AI whether your plan is good, it often finds reasons to agree, especially when you mention which option you like. That is fine for a recipe and a problem for a decision about money, time or a title you will put your name on. The fix is a different question. Instead of "is this good?", ask it to argue the other side, imagine the failure, list its guesses or doubt itself.

These five prompts do that. Each one comes with a real example of what changed.

Words used here

prompt: the message you send to the AI.

Before you start

An AI chat app, for example Claude, ChatGPT or Gemini. The prompts are plain text and work in all of them.
A decision you have not made yet, written in two or three sentences with real numbers in them.

How to use them

Copy the prompt, replace the line in square brackets with your decision and your situation, and send it in a new chat, because an old chat already knows what you think. Add facts such as followers, savings or the deadline. The more you add, the more specific the answer. If you leave them out, the assistant asks for them instead of answering.

1. The other side

What this is for: making the assistant argue against your plan, the way a friend who has seen it fail would.

I am about to make a decision and I want the other side before I do. Argue against it as strongly as you honestly can, the way a smart person would who has seen this go wrong before. Use the facts I gave you, and mark anything you add from general knowledge as general knowledge. Then tell me which of your own arguments is the weakest. Do not praise my plan and do not soften anything, but if you cannot find a real objection, say so instead of inventing one. My decision and situation: [PASTE YOUR DECISION AND YOUR SITUATION HERE: the numbers, the deadline, what you have already tried]

Example. I asked which of two titles for this guide to use, and added "I like A".

Without the prompt"Go with A. It's concrete (five prompts, a clear deliverable), and it names a frustration beginners have already felt."

With the prompt"Recommendation: use B, or B with A's concrete hook as a subtitle." It gave five arguments against A, including "Your preference isn't evidence. You told me you like A, but nothing you've told me shows A performs better with your audience."

2. Six months later

What this is for: finding the ways a plan fails, with a sign to watch for and a cheap fix for each.

Imagine it is six months from now and the decision below turned out badly. Write down the 4 most likely reasons, the most likely first. For each one, give me one early warning sign I could notice in the first month and one cheap thing I could do now to reduce the risk. Finish by telling me whether any of them is serious enough that I should change or delay the plan. My decision and situation: [PASTE YOUR DECISION AND YOUR SITUATION HERE: the numbers, the deadline, what you have already tried]

Example. I asked about launching a $9 a month membership with 2,400 followers and 150 people who open my weekly email.

Without the prompt"Yes, but test it before you commit." It counted paying members from the follower count and got "roughly 15-40 members, so about $135-360/month".

With the promptIt counted from the 150 people who open the email: "5-8 members, or roughly $45-70 a month". It listed four ways the plan fails, each with a first-month warning sign, and ended with a go or no-go number of 15 paying members.

3. What must be true

What this is for: showing which parts of your decision are facts you gave and which are guesses the assistant filled in.

Before you give me any opinion, list what has to be true for this decision to be a good one. Split the list into things I told you and things you are assuming. Pick the two assumptions that would change your answer if they were wrong, and for each one tell me the cheapest way to check it in under a week. If one missing fact matters more than all the others, ask me that one question at the end, and only then give a recommendation. My decision and situation: [PASTE YOUR DECISION AND YOUR SITUATION HERE: the numbers, the deadline, what you have already tried]

Example. I wrote that I had decided to quit my job next month to run my channel full time, with 2,400 followers, no income and four months of savings, and asked it to confirm.

Without the prompt"I can't confirm it, and with these numbers I'd advise against quitting next month." Four reasons and a list of conditions for quitting later.

With the promptA list in two parts, "Things you told me" and "Things I am assuming". It then asked one question: "Has anyone, whether a viewer, a company, or a client, ever offered or said they'd pay you for something connected to the channel, and if so, what for and how much?"

It waits for your answer before it recommends, so you get the one question that matters most.

4. Check yourself

What this is for: asking the assistant to doubt its own answer, so you see which parts it is guessing.

Answer the question below. Then, in a separate section, read your own answer like a skeptic: which statements are you least sure of, which ones did you say with more confidence than the facts allow, and what would you say if I told you I disagree? Finally, rewrite the answer with those fixes. If nothing needs to change, say "I stand by it" and explain why. My question: [PASTE YOUR QUESTION HERE]

Example. The title question again.

Without the prompt"Go with A." Each of three runs added only a small warning that the title "slightly overpromises".

With the promptIt answered "Use Option A", then under "Said with too much confidence" it wrote: ""Names a problem they've already felt" was stated as fact. It's an assumption." The rewritten answer became "Go with A, but treat it as a reasonable bet rather than a clear winner."

This prompt keeps the answer and adds honest doubt, so you see which parts are guesses.

5. No peeking

What this is for: removing your own preference from the question, so the assistant cannot lean toward it.

Ignore which option I said I prefer, treat my preference as unknown. For each option, give the best case its own champion would make, in about three sentences. Then choose one option by a single criterion that you name, and tell me what evidence would flip your choice. If my situation is not a choice between options, treat "do it" and "do not do it" as the two options. My decision and situation: [PASTE YOUR DECISION AND YOUR SITUATION HERE: the numbers, the deadline, what you have already tried]

Example. The title question.

Without the prompt"Go with A", while it knew that I like A.

With the promptIt wrote the best case for each title and chose "B, judged by a single criterion: how clearly a beginner who has never noticed AI sycophancy understands what they'll get." Then it listed what would flip the choice to A. Sycophancy means an AI agreeing with you to please you.

The list of what would flip the choice is the most useful part, because it turns an opinion into something you can check.

Which one to use when

Prompt	Use it when	What you get
1. The other side	You already lean one way and want the best case against it.	The strongest objections, and the weakest one marked.
2. Six months later	You need things to watch before a launch, a purchase or a big change.	Four likely problems, each with a warning sign and a cheap fix.
3. What must be true	You feel sure and cannot say why.	Your facts and guesses in two lists, and the one question that matters.
4. Check yourself	You have an answer you like and want to know which parts are guesses.	The same answer with honest doubt added where it is needed.
5. No peeking	You choose between two things and have said or hinted which one you like.	A fair case for each option and what would flip the choice.

If something looks off

It asks me what my decision is

You sent the prompt without replacing the line in square brackets, or the prompt arrived without the second half. Paste the whole prompt including your decision in one message.

It gives me a long list of objections

Prompts 1 and 2 ask for objections on purpose. Read the part where it names its weakest argument, and use the rest as a checklist. Prompt 5 asks for the best case on both sides instead.

It still agrees with me

Start a new chat, remove any line that says which option you like, and use prompt 5. If it still agrees, ask it to list what would change its mind and check those things yourself.

How to get the best result

Give real numbers. Your audience, your costs, the deadline and what you already tried make the answer specific to you.
One decision per chat. Start a new chat for each decision, so the assistant works only with what you wrote.
Check the numbers it adds. Anything it marks as general knowledge is a good thing to compare with your own data.
Use two prompts together. Run prompt 3 first to see what must be true, then prompt 2 to find the early warning signs.
Big decision? Take the list of objections and checks to someone who knows your field. It gives the conversation a head start.

Want more prompts like this? Follow my content and check my other posts.

Get the next guide when it is ready.

One email a week at most, and you can leave whenever you like.