This week I had two calls with Nicolas Boucher.
He’s now at 3.7M+ followers, with a $2M+ business that’s on track to do a lot more this year. He was super excited as he’d spent the earlier part of the week using his digital clone in ChatGPT as a thought partner to inspire some amazing ideas about his business.
All in an effort to improve the value he delivers through his AI Finance Club.
But what I found interesting was the way he uses AI. He teaches AI in finance yes, but he’s getting a massive amount of value from a few tools used well. Against some companies that would have developed a 37 agent swarm to do the same thing (and probably worse).
So here’s his exact setup (and how it varies against my own).
Part 1 - The Knowledge
Most of his setup is a project in ChatGPT.
In both ChatGPT and Claude you can create a Project (a workspace with background knowledge, files, and instructions that every chat uses for context without having to re-explain).
His projects already know:
What his company does
Who's on his team (so when he types a teammate's name, it knows their role without him explaining)
How he likes work formatted
By the way, this is exactly what Estée Lauder did at scale - they built 240+ custom GPTs loaded with 75+ years of consumer and clinical data so their teams can query their own knowledge (not the internet).
My own AI workflow is a combo of Claude and Notion (Notion acts as my digital brain as all my notes are already there - I can tag different pages, and combine pages for better context, or chain multiple prompts together - cool eh?)
Part 2 - The Interview
If there’s one stupidly simple change you make today, it’s this.
He tells the AI to ask him questions before it does anything.
On our call he described setting up the AI to act like a coach:
“A coach asks questions before giving instructions, but still does its job, which is to find the solution. If you don't tell it that, it "just acts like Google": you ask a question, it gives you an answer, even when it didn't have enough information to give a good one.”
And crucially — he gives open context, not leading context.
Open is when you deliberately don’t tell it all of your thinking because you want a useful answer that might challenge your point of view. Leading context is if you tell it what you’re already thinking, which makes the AI more likely to give a biased response. You’ve probably experienced this when AI ends up just agreeing with you.
His example: if you say "I'm thinking about hiring someone," the AI just produces a lot of text telling you how to hire someone. But if you stay wide and open, it surfaces options you'd never have thought of - because you didn't bias it.
Dan Koe (7-figure solo creator) does this too. His "two-phase" prompt:
Phase 1 - the AI interviews YOU for context.
Phase 2 - it generates from your answers, so it sounds like you, not a robot.
His prompt even instructs the model to be "purely observational 'clear eyes' that does not give me the exact answer but guides me to it" and to ask one question at a time.
The last time I got AI to interview me was to pull out stories, methods, and wins from my previous work to inspire content ideas.
But most of the time I add this line to all of my prompts:
“Ask me any questions you need before we start.”
Part 3 - The Tutor
There's a difference between using AI to think for you and using it to think with you.
Nvidia CEO Jensen Huang uses ChatGPT, Gemini and Perplexity "as a tutor every day." His method is to: "Start by explaining it to me like I'm a 12-year-old, and then work your way up into a doctorate-level over time."
Nicolas does something similar. He uses it to frame and advance his own half-formed ideas. His words: "it's starting to frame your ideas in a way that's a bit more advanced than what you had." And when it tried giving definitive answers too early he’d say: "No, I don't want to go there - spend more time on the options."
I often use ChatGPT for personal learning. Although right now I just feel like the last 3 questions I asked it were about how long fish will last in the fridge.
I did actually set up a Claude project for training my puppy Luna. But there are some things that even a super advanced AI cannot help with XD.
I do quite often use Claude with my kids, I ask them what they want to learn about, and then I turn it into a fun bedtime story where they learn whilst I speak.
Part 4 - The ‘Stack’
I hate the word ‘stack’ but for lack of a better word, let’s use it here:
Nicolas:
ChatGPT: strategic thinking, brainstorming, and his digital clone work (plus also images are really good).
Claude: Finance work, analysis, and heavier building (like modelling and more complex rule based tasks).
Gemini: For deep research (both web and Google Workspace files)
Me:
Claude: For pretty much everything. I have lots of projects, and a lot of the time I use it for analysis (subject line analysis, copywriting analysis) - plus to spit out a first draft so I’m not looking at a blank page (like I’ve just done for this newsletter.
Gemini: Image generation - I used it for the thumbnails for these very newsletters (it’s a Gem with my image style pre-loaded as knowledge, like a media kit)
Notion: I use this for research (like finding some of the above quotes you’ve seen) and to help draft social media posts, as all of my posts are already there in a Notion database. I find working in Notion is super fast, especially for content production (plus you can even choose which AI model you want).
Don’t overcomplicate things
Strip away the LinkedIn hype and a $2M creator's AI workflow comes down to 4 simple principles.
Build a Project that already knows your world
Make AI interview you before it generates anything
Don’t overshare your thought process so you keep the AI answers unbiased
Use it as a tutor, not an oracle
Plus, don’t overcomplicate things.
If one tool is really sh** at something, use another one.
There’s no pain in having one ChatGPT tab and one Claude tab open.
Pick your battles, but cramming everything you do into one tool just sometimes aint’ gonna cut it.
Until next time.
Adam