Give It Your Own Last 30 Days

Open your calendar, scroll back to this day last month, and then come forward to today. Every meeting you moved is there in order, and so is every evening you blocked out and then gave away. And you have never once read the month as a whole, because nobody does.

Your sent folder knows who you wrote to and how long you took to reply, and your bank knows where the money went. Your phone knows how far you walked, and the notes app knows what was on your mind at midnight. All of this is spread across six apps, and the one time you tried, you got as far as a spreadsheet and gave up.

AI does this well, and you couldn't do it before. It can read a month of your records and say what it sees. Spotting patterns across lots of small records is what an AI model is good at. A model is the system behind the chat. A person is bad at this job.

Memory holds the week that felt busy, but the calendar knows whether it was. You may think you reply fast, but the sent folder has the times. Thirty days is long enough for a shape to appear, and short enough to export without losing the will.

But point AI the wrong way and it finds things that aren't there, because if you ask AI for patterns, it will always find some. A line like "You're most productive on Tuesdays" might be a real rhythm, or it might be two quiet Tuesdays in a row.

AI won't know the difference unless you make it say how sure it is. And you won't know unless you push back with what the data can't see, for example that you were off sick on one of those Tuesdays.

A month is a small amount of data, and AI will happily find a trend in four entries. Let AI look first, and then argue with it.

One month, one source

1. Export one thing. Don't export all six, just one. That could be your calendar as a list, or your sent emails with the subject, who it went to and the time, without the message itself. Or it could be your spending as a statement, or your to-do app's done list.

   Most apps can save your data as a file, and if yours can't, AI can read screenshots. What you paste is covered by the settings of whatever AI tool you're using. So check those settings before you paste a month of anything, especially spending or messages. Remove names and account details you don't need.

   The patterns are in the shape of the month, so the details can stay with you.

2. Ask what AI notices, and how sure it is.

   Here's a month of my [calendar / sent emails / spending]. Tell me five things you notice about how I spent the month. For each one, say how confident you are and what in the data makes you think it. Flag any that could just be coincidence in a small sample. Don't give me advice yet. Don't flatter me. Just the patterns.
   

3. Argue with the list. Some things only you know, like "That week was a conference" or "I was ill on the 14th". Give AI the context and ask again.

   Here's what you're missing. [the context]. Given that, which of your five observations still hold? Drop the ones that don't and tell me why.
   

   What's left is usually two or three things, and they're usually true. Skip the argument and you'll end up reorganising your week around a coincidence.

4. Keep one. Pick the single pattern you'd change something about, then decide what to change. For example, "I'm spending every Monday morning on email I could batch," or "I haven't written to anyone outside work in a month."

   Then put the export away, because you'll do this again in a few months and you'll want to compare.

If you want to make this regular without the fuss, set up a Project. A Project is a space in your AI app that remembers your standing instructions. Give it the instruction "when I paste a month of calendar data, run the pattern check and the confidence flags", and the second step becomes one paste.

Perhaps you don't want a mirror at all. Some people find a month of their own records bleak, and a tool that's confident about what it sees in you can land harder than a spreadsheet. If that's you, start with the spending or the calendar and leave the messages alone. Treat whatever comes back as a look, not a verdict.

AI counts the Tuesdays. You know which one you spent ill.

---

Credit goes to Will Francis, who has shown the "feed it your own analytics and ask what it notices" move in his own videos. The method and the confidence-flag step here are mine.