Everyone tells you to audit your subscriptions. Nobody tells you how, so nobody does it.
Here's the how. Five minutes, tonight, on whatever export your bank gives you.
Paste your last three months of card transactions into AI with this:
"Here are three months of my card transactions. Find every recurring charge that appears in all three months. For each one, list the merchant, the amount, and how many of the three months it appears in. Flag anything that looks like a subscription. Then separately list any charge that appears in only one or two months, since those are the ones I've likely forgotten I signed up for or already cancelled. Do not guess at anything not in the data."
That last line matters. Without it the model starts inventing plausible subscriptions you don't have. You want it reading, not imagining.
Here's what a result tends to look like. Say the list comes back with eleven recurring charges. Nine you recognize instantly. But two are the interesting ones: a $9 charge you stopped thinking about a year ago, and a tool at $19 a month that you replaced back in spring and never cancelled.
That's the whole point of the exercise. Not willpower. Not discipline. Just getting every charge onto one screen where your brain can finally see them side by side. The leaks were never hidden. They were just never in a list together.
Now here's the part I want to be straight about.
The prompt finds the list. That's all it does.
It won't tell you that the $19 tool has a free alternative that does the same job. It won't tell you your $70 plan drops to $55 for the exact features you use. It won't tell you which of those eleven to keep and which to kill. And it definitely won't still be right in three months, when the prices move and the free tier you relied on adds a paywall.
Finding the leak is five minutes. Knowing what to do with each line is the actual work. That part is still on you.
Run the prompt tonight anyway. Even the raw list is worth the five minutes.
Stay sharp.
— Chris
P.S. Run it and reply with how many recurring charges it found. I'm genuinely curious what the spread looks like across people. The number is usually higher than the guess.