5 Lessons Learned After Exchanging 5.7 Million Words with ChatGPT Over the Last 6 Months
AI hasn’t replaced us.
But it has made us much harder to replace.
In the last 6 months, we’ve exchanged over 5.7 million words with ChatGPT.
That’s a lot of prompts, experiments, and “wait… can it actually do that?” moments.
Some of those words were wasted. Some were brilliant. All of them taught us a lot about how to use AI effectively in a real business setting.
Here are the 5 biggest lessons we’ve learned:
1. Use deep research, then consolidate
If you just ask AI something like “what’s the best CRM for agencies?” you’ll get the same generic surface-level answers everyone else does.
But if you go deep first — gather competitor data, build your own comparison criteria, or pull industry-specific insights — then feed that into ChatGPT? That’s when it starts to feel like you’ve got an extra pair of very smart (very fast) hands on your team.
AI is at its best when it’s synthesizing your context — not when it’s guessing in the dark.
2. Test its boundaries to see its potential
The most interesting outputs we’ve had weren’t from “normal” prompts.
They came from pushing the model to its limits.
Can it roleplay as a customer during a mock sales call?
Can it take raw transcript data and extract emotional cues?
Can it map out an automation strategy based on our tech stack?
The point isn’t that it nails it every time.
It’s that by stretching its edges, you quickly find the “hidden” use cases most people never uncover.
If you only use AI for obvious tasks, you’ll only get obvious results.
3. Give it examples of your ideal output structure
Prompts are instructions. Examples are cheat codes.
When we started giving ChatGPT specific examples of what we wanted — an email in our tone, a report using our formatting, a post like one of ours — the quality of its outputs leveled up dramatically.
“Write us an email” will get you an email.
“Write us an email that looks like this one we sent last week, but tailored to this new client persona” will get you something that’s actually useful.
Show it the shape of the thing you want, and you’ll get 10x better results.
4. Phrase questions in as much of an unbiased way as possible
This one’s subtle but huge: leading prompts lead to limited answers.
If you ask, “Why is X the best choice?”, you’ll get a confirmation-biased list of reasons why X is the best choice.
But if you ask, “Compare X with Y and Z. What are the trade-offs for each?”, you’ll get something way more nuanced and decision-ready.
When you strip your assumptions out of the prompt, you open the door to surprising insights. And those are often the ones you actually need.
5. Keep experimenting with different models
ChatGPT isn’t one monolithic AI. It’s a collection of models, each with its own quirks.
Some are better at creative ideation.
Some are sharper at analytical tasks.
Some are faster and cheaper (and that matters too).
Switching between them feels like getting multiple expert opinions on the same problem. It also stops you from getting too comfortable with one “AI voice,” which can be a trap in itself.
So, what’s the big takeaway?
After 5.7 million words, here’s what we know for sure:
AI won’t do your job for you.
But it can make you dramatically better at doing it — if you learn how to use it properly.
Think of it less like a magic button, more like an amplifier. The more clarity, context, and creativity you bring, the more it gives back.
And that’s what makes it worth every one of those 5.7 million words.