The 6-Prompt System That Tells You Whether Your Doubts Are Real
How to Build an AI Assumptions Log to Stop Second-Guessing Business Decisions
Second-guessing is an information problem, not a confidence problem.
When something goes sideways (a business bet taking longer than expected, a career move that feels less certain three months in), most people can’t remember exactly what they believed when they made the decision. So they relitigate the whole thing. Was it the right call? Should I have known better? They’re not questioning the decision so much as the reasoning behind it, and that reasoning has gone fuzzy.
The fix isn’t more conviction. It’s a record. Specifically, a log of the assumptions your decisions were built on, updated as evidence comes in, readable by an AI that can tell you whether your current doubt is signal or noise.
That’s what this article builds.
What You’ll Get
By the end you’ll have:
An assumptions log template structured for AI to reason about, not just store
A prompt that reverse-engineers the assumptions underneath decisions you’ve already made
A maintenance prompt that maps new evidence to existing beliefs and updates their status
A second-guess audit prompt that tells you whether your doubt is real or just anxiety
A stale assumptions scanner that finds decisions still in force built on beliefs that no longer hold
A pre-decision capture prompt so future bets start with a clean record
Setup takes about 25 minutes the first time. After that it’s a few minutes when something changes or a doubt surfaces.
Two People Who Need This
Marcus runs a small SaaS business. Six months ago he pivoted from a consumer product to B2B, based on a set of beliefs about enterprise willingness to pay, sales cycle length and his team’s ability to close. Things are moving, but slowly. He second-guesses the pivot almost weekly.
Diane left a senior role at a large firm to join a 40-person company as Head of Strategy, trading title and salary for what she believed would be faster learning, more autonomy and a clearer path to the C-suite. Eight months in she’s not sure. The learning is real but the path feels murkier than expected.
Neither of them is in a crisis. Both are in the same trap: they can’t remember precisely what they believed when they decided, so every bad week feels like evidence they were wrong. The assumptions log doesn’t tell them whether they made the right call. It tells them whether their doubts are based on something that’s actually changed.
Prompt 1: Build the Assumptions Log Template
What it does: Generates a structured log format that an AI can read, compare against new evidence, and reason about over time.
When to use it: Once, at setup. This becomes the document you maintain and reference going forward.
The Prompt:
Create an assumptions log template for someone tracking the beliefs
behind a major business or career decision.
Each assumption entry should capture:
- The assumption itself, stated as a plain belief
(e.g. "Enterprise buyers in this space will pay £500/month")
- The decision it underpinned
- When it was logged
- Current status: HOLDING, WEAKENED, or INVALIDATED
- Evidence that changed the status (leave blank at start)
Format requirements:
- Plain text, easy to paste and update
- One section per major decision or bet
- A summary line at the top of each section showing
how many assumptions are HOLDING / WEAKENED / INVALIDATED
- Simple enough that updating it takes under 5 minutes
Include two placeholder examples — one business bet,
one career bet — to show the format in use.How to use it:
Run the prompt and save the output as a document in your Claude Project, or paste it into a note you can access quickly.
The placeholder examples show you the format. Replace them with your own in the next step.
Every future prompt in this sequence references this document.
Example input: No additional input needed. Run as written.
What you’ll get: A clean log structure with a summary header per decision, individual assumption entries with status fields, and two worked examples showing what a live log looks like. Something like:
ASSUMPTIONS LOG
===============
DECISION: [Name of bet or decision]
Date decided: [DATE]
Status summary: 4 HOLDING / 1 WEAKENED / 0 INVALIDATED
Assumption 1: [The belief]
Decision it underpins: [What you chose based on this]
Logged: [DATE]
Status: HOLDING
Evidence: —
Assumption 2: [The belief]
Decision it underpins: [What you chose based on this]
Logged: [DATE]
Status: WEAKENED
Evidence: [What changed and when]Pro tip: Keep the assumptions short and falsifiable. “The market is big enough” is too vague to update. “There are at least 5,000 businesses in the UK spending over £1k/month on this problem” is something that can be HOLDING or INVALIDATED based on actual evidence.
Prompt 2: Extract Assumptions From a Decision You’ve Already Made
What it does: Takes an existing decision and reverse-engineers the implicit beliefs that were underneath it, so you can log them even if you never wrote them down at the time.
When to use it: For any decision already in flight. Most people won’t be starting from scratch.
The Prompt:
I made a significant decision recently. I didn't explicitly log my
assumptions at the time, but I want to reconstruct them now.
Here's the decision and my best recollection of the reasoning:
[DESCRIBE THE DECISION AND WHY YOU MADE IT — as much detail as you
can remember, including what you were hoping for and what you were
worried about]
Your job:
1. Identify the key assumptions I was implicitly making. Look for
beliefs about: the market or environment, other people's behaviour,
my own capabilities, timing, and what would happen if I was right.
2. State each assumption plainly, as a specific falsifiable belief.
3. For each one, ask me: does this still feel true, partially true,
or no longer true? I'll update the statuses myself, but flag
anything where you'd want more information before classifying it.
Format the output ready to paste into my assumptions log.How to use it:
Write your description of the decision messily. Don’t structure it. The AI will do the extraction.
Answer the status questions honestly. This is where the value is. Not in the format, but in being forced to say “actually, I no longer believe that.”
Paste the output into your log document with statuses filled in.
Example input:
“Six months ago I pivoted my SaaS from B2C to B2B. My thinking was that consumers weren’t willing to pay enough and that businesses would value the workflow benefits more. I thought we could close deals in 4-6 weeks and that my existing network would get us the first 10 customers without needing a dedicated sales hire. I was also betting that our small team could handle the added complexity of enterprise support without burning out.”
What you’ll get: Six to eight explicit assumptions, pulled from what was implicit in the description. Something like: “Enterprise buyers in this category will pay a meaningful premium over consumer pricing” / “Initial sales cycles will be 4-6 weeks” / “Founder network is sufficient to generate the first 10 customers” / “The team can absorb enterprise support demands at current headcount.” Each one ready to be marked HOLDING, WEAKENED or INVALIDATED.
Pro tip: The assumptions you resist logging are usually the ones most worth logging. If you find yourself thinking “well, that one’s obvious,” write it down. Obvious assumptions are the ones that quietly invalidate without anyone noticing.
You now have a Brand Profile and a defined Audience Persona, the ‘Who’ and ‘To Whom’ of your business. But if you stop here, Claude will still default to its own robotic prose because it doesn’t have your specific DNA yet.
To turn these profiles into a high-converting output engine, you need the Execution Layer.
Paid subscribers get the remaining 6 prompts to finalize their Skills File:
The Voice & Tone Extractor: Stop ‘prompting’ for style and start extracting it from your best work.
The Platform Rulebook: Automated constraints for LinkedIn, Email, and Ads so you never have to fix a 600-word ‘short’ email again.
The Self-Revision Checklist: A recursive loop that forces Claude to edit its own ‘AI-isms’ before you ever see a draft.
The Master Template: A copy-paste Markdown file to set up your Claude Project in 60 seconds.
Upgrade to unlock the full system and stop billing the time you didn’t save.