AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people meet "AI made me rich" stories as a screenshot and a headline, then copy the most visible detail: the tech stack, the launch tactic, the tool. It feels like learning because the numbers are real somewhere. But the visible detail is rarely what produced the money, and you end up rebuilding someone's file structure instead of their position.
The rule: before you copy anything from a success story, put it through the Proof Ledger. Write down what the receipt actually measures, what came before it, and which part depended on things you do not have. Copy only the rows that survive.
This guide runs the ledger on two of the best-documented solo AI businesses, then shows you how to use it on the next story that lands in your feed.
Both founders published their own figures. That makes them unusual, and worth reading properly. Figures are in US dollars as the founders reported them; I have not converted them, because the exchange rate on the day would add a false precision.
| Photo AI (Pieter Levels) | Headlime (Danny Postma) | |
|---|---|---|
| What it is | AI photo generator | AI copywriting tool |
| The headline claim | One index.php file of 40,870 lines, $105,000/month revenue, $80,000/month profit |
Built and sold for a seven-figure sum in about eight months |
| Where it comes from | Levels' own post, 6 March 2026 | Postma's own interview with Failory, plus his acquisition announcement on X |
| What it measures | Monthly revenue and monthly profit at one point in time | A sale price ("7 figure sum"), not a disclosed exact number |
| Date of the event | March 2026 | Sold March 2021, to the company then called Conversion.ai/Jarvis (now Jasper) |
Two things to notice straight away. Levels reports a single month, not a trend; his own earlier post (July 2023) put Photo AI at almost 14,000 lines and $61,808 a month. And the "about $1M" figure for Headlime is the community's shorthand. Postma's own words are "a 7 figure sum", which is a range, not a number.
No, and this is the row most retellings drop.
So both stories are the visible end of a long run of attempts, an existing audience, and good timing with a new model. None of that shows up in a screenshot of a revenue number.
The trap is copying the surprising detail instead of the load-bearing one.
| Surprising detail people copy | What was actually doing the work |
|---|---|
| "One giant PHP file" | Fast shipping by one person who already knew the stack well, plus distribution through a large following |
| "Built in eight months" | Years of earlier products, a validated idea from his own book, and a launch timed with GPT-3 |
| "AI did it" | AI was the product's engine; the founders still did the selling, pricing and support |
| "Sold for $1M" | A buyer that wanted the product and customers in a hot market, in 2021 |
Quieter traps: a single month's revenue says nothing about churn or refunds, a sale price says nothing about deal terms, and you only hear from the founders who won.
Use one row per claim. Fill it in before you decide anything.
Suppose you run a small bookkeeping practice in Leeds and a post claims "solo founder built an AI receipts app to £20K a month in 90 days". Here is how the ledger might look. Every detail below is invented for illustration.
| Row | Entry |
|---|---|
| Claim | £20K/month within 90 days |
| Source | A repost on a business page; no link to the founder |
| Measure | Unclear; could be revenue or bookings |
| Date | Not stated |
| What came before | Unknown |
| Dependencies | Probably an AI model API and app-store approval |
| Copyable move | Turning a repeated client chore into a small paid tool |
| Verdict | Skip the numbers; keep the copyable move as an idea to test with three existing clients |
A weak source does not make the idea worthless, only the numbers.
Paste a success story into Claude or ChatGPT with this prompt. Fill in the story text and your own situation.
You are a sceptical analyst helping a small-business owner decide what, if anything, to copy from a published success story.
Story text or link summary: [PASTE_STORY]
My situation: [YOUR_BUSINESS, SKILLS, AUDIENCE_SIZE, BUDGET, HOURS_PER_WEEK]
Goal: fill in a Proof Ledger so I can see which parts of the story are evidence and which are noise.
Steps:
1. Split the story into separate factual claims, one per row.
2. For each claim, record: source (first-party, founder-controlled dashboard, or third-party retelling), what it measures (revenue, profit, recurring revenue, sale price, users), and the date it was true.
3. List what came before the result: earlier products, audience, money, skills, timing.
4. List dependencies the founder did not control.
5. Name at most two moves I could realistically repeat given my situation, and one thing I should not copy.
Output: a table with columns Claim | Source | Measure | Date | What came before | Dependencies | Verdict, followed by the copyable moves and the "do not copy" item as short bullets.
Rules:
- Mark anything not stated in the story as "Not stated". Do not fill gaps with guesses or typical figures.
- Do not convert currencies or project future revenue.
- If my situation details are missing, ask me for them before giving verdicts.
Before answering, check: does every number in your table appear in the story text? Remove any that do not.
Fill in: the story, and an honest description of your own starting point.
What survives the ledger from these two stories is modest and useful:
What it will not do: give you their audience, their back catalogue, or 2021's market for AI copywriting tools.