Don't Fool Yourself: The AI Prompt That Challenges Your Own Thinking

The Briefing

GPT-5.4 mini and nano just dropped into the API. OpenAI released two new models this week. Mini brings near-flagship performance at a fraction of the cost. Nano is built for high-volume, simple tasks where speed is everything. If you're running any API automations — data pulls, content generation, classification — these are worth swapping in. Faster output, lower token cost, same quality on most marketing tasks.

AI design tools are now mainstream, not experimental. 75% of marketers are using AI for image creation in 2026. Teams report 40–70% cuts in creative production time. Midjourney v7 and Adobe Firefly 3 are leading the pack — Midjourney for hero visuals, Firefly for legally safe commercial assets. If you're still outsourcing every image, the math has changed.

OpenAI started running ads inside ChatGPT. This matters for marketers. Your customers are now asking: is this recommendation organic or paid? Authenticity used to be assumed in AI responses. It isn't anymore. Something to factor into how you position yourself as a practitioner in this space — and another reason to have your own voice, not just AI's.

Unbounce Smart Traffic is reporting 30% conversion lifts without manual A/B tests. Their AI auto-routes each visitor to whichever page variant is most likely to convert them — continuously, based on live behavior data. Not a split test you run and evaluate. A system that improves itself. If you're still manually testing one variant at a time, this is the direction the whole industry is moving.

The Build

Happy April Fools' Day. And today's topic is appropriate: the biggest fool in any business isn't the market, the algorithm, or the economy. It's you — and the stories you tell yourself about your own data.

Clients closed out a solid month. Low-ticket front-end sales pushed list volume, the higher-ticket conversions followed, and the numbers improved month-on-month. Feels good. But "feels good" is exactly where the trouble starts.

This week I ran the same API data pull multiple times with small variations. Different angles on the same conversion numbers. Looking for what was actually driving the result versus what I wanted to be driving it.

Here's the pattern I see constantly — in my own work and in the work of people I coach: you get a result, you build a story around it, and then every data point you look at after that confirms the story you already decided on.

The AI makes this worse, not better. Ask Claude "why are my conversions improving?" and it will find you a reason.

Ask it the right way and it will find you the truth. Those are two very different things.

The fix is this week's file: thinking-partner-prompt.md

This isn't a marketing framework. It's a meta-prompt. You load it before any analysis session and it changes the entire dynamic. Instead of you presenting a narrative and Claude confirming it, you're forced through a six-step process that's designed to surface what you're missing:

Step 1 — What are you actually saying vs. what you think you're saying?

Step 2 — Where is your reasoning broken?

Step 3 — What are you avoiding, and what is it costing you?

Step 4 — What would someone who's already where you want to be do differently?

Step 5 — What's the precise action plan — including what to stop?

Step 6 — What's the one question you're clearly avoiding?

The ground rules are as important as the steps. No opening with praise. No softening critiques. If your plan is solid — stress-test it harder.

The prompt literally says: "You are not my cheerleader. You are not my yes-man."

I used it three times this week.

First, on the conversion data. I fed it the month-end numbers and told it I thought improvement came from better front-end targeting. It pushed back: you're measuring yourself against your own history, not against what's possible. Your numbers are up — but there are three friction points in your funnel you haven't touched. Here's what each one is costing you per month.

My story was true. It just wasn't the full picture.

Second, on page design. I've been moving pages from stale and forgettable to pages that actually make someone want to stay. Imagery placement matters more than most people realise — the wrong image creates a split second of cognitive dissonance between the headline and the visual, and that's enough to kill conversion. I asked the thinking partner to evaluate a layout I liked. It told me the imagery I chose prioritised aesthetics over conversion speed. It was right. I'd been fooling myself that the two were compatible in that particular layout.

Third, on a pricing question about the front-end offer. I thought I was making a strategic move. The thinking partner pointed out I was optimising the visible thing to avoid dealing with the harder question underneath it.

Classic April Fools move — on yourself.

Here's the prompt structure that kicked off each session:

You are my brutally honest thinking partner. Use the six-step framework below.

[Paste the full thinking-partner-prompt.md here]

Here's my situation: [Describe what you're deciding, analysing, or stuck on]

Do not open with praise. Do not soften the critique.
Tell me what I'm not seeing — especially where my own bias might be colouring the data.

What worked: Using it AFTER forming an initial view. Get the data, build your interpretation, then run it through the thinking partner to stress-test. The sequence matters — bring it in too early and it kills ideas before they have legs. Bring it in after you've formed a view and it sharpens everything, including the parts of your thinking you didn't know were soft.

What didn't work: Using it on a decision I was already emotionally committed to. When you're looking for validation, this prompt won't give it. That's the point — but it means you have to be honest with yourself about why you're opening the session. "Am I looking for insight or am I looking for agreement?" Answer that before you start.

The Play

Before your next big marketing decision — a funnel change, a price test, a creative overhaul, a new offer — run it through the thinking-partner prompt first. Not for ideas. For catching the story you're already telling yourself.

Two things to watch for in particular.

Step 3: what are you avoiding, and what is it costing you? This one exposes the expensive habits hiding behind "I'll get to that later."

And Step 6: the uncomfortable question at the end. That's the one that actually changes the decision.

Critical rule from this week's API work: you must know the exact output you're looking for before you make the request. Vague in, vague out.

If you ask Claude to "have a look at my data and tell me what you see," it will construct a narrative that sounds smart and confirms your existing direction. If you ask it to find the assumption you're most likely wrong about — you get something completely different.

You are my brutally honest thinking partner. Your job is to make my thinking sharper and my blind spots visible.

You are not my cheerleader. You are not my yes-man.

Here is my situation and my current plan:
[Describe your marketing decision — what you're planning, why, and what result you're expecting]

Specifically: challenge the interpretation I've put on the data. Where is my own bias most likely distorting what I'm seeing? Follow your six-step framework. End with the one question I'm clearly avoiding.

One more thing worth building: a backlog. Every sharp insight from a thinking-partner session — save it. The same way a library of past design decisions improves the quality of your creative output, a library of past thinking-partner sessions shows you your own recurring blind spots. Patterns emerge. You start to see the exact ways you consistently fool yourself. That's where the real leverage is.

Try this today. Take one decision you've been sitting on for more than a week. The one you've been calling "thinking about it." Run it through the prompt. See what you've been avoiding.

Download thinking-partner-prompt.md