Knowledge base › Tools

Ask someone how the last three months went and their mood today answers, not the three months. That is why we miss our own patterns: from the inside, every situation feels unique, and memory edits after the fact. A dated journal and an AI that can lay many entries side by side can show you things you would not spot alone — if you know which patterns to look for and what to do with one once you find it. Here are the five most common pattern types, a 90-day plan, a two-week test, and an honest look at where the AI gets it wrong.

Why you can't see your own patterns

A pattern is a pattern because it repeats. But repetition is almost impossible to notice from the inside, because every single instance feels unique. Tuesday's argument had its reasons, so did Thursday's exhaustion, so did Sunday's dread. One by one, each makes sense. Only side by side would you see that it's the same story on a different set.

The second reason is memory. It is not an archive; it is an editor, and it selects according to your current mood. Psychologists call this mood-congruent memory: in a low mood, bad memories come to mind more easily; in a good mood, the good ones do. That's why in a hard week you say “it's always like this”, and in a good week “honestly, everything's fine”. Neither is a lie. Neither is accurate.

The third reason is sheer volume. Ninety days means ninety mornings, ninety evenings and hundreds of small decisions and conversations. Nobody holds that in their head. Most people remember a handful of standout days from their last quarter — the ordinary days, where patterns actually live, blur together.

So you need something that doesn't remember by mood: a dated journal that someone — you, or an AI — reads all at once, entries side by side.

You don't miss your patterns because you aren't paying attention. You miss them because you only ever see one day of them at a time.

Five pattern types that show up within 90 days

“Pattern” sounds vague. In practice there are five distinct kinds. They are worth knowing, because each needs different data and each calls for a different response.

• Time patterns. When does it happen? Time of day, day of the week, point in the month. For example: most of your anxious entries are written on Sunday evenings; your energy reliably runs out around 3 p.m.; in the last week of the month you write more tensely about money. No deep analysis needed — just dates and consistency.
• Cause-and-effect — or rather, co-occurrence — patterns. What comes before it? The classic is poor sleep followed by a shorter fuse the next day. But it also covers always feeling drained after one particular meeting, or writing more calmly on days you exercised. Important: what a journal shows is co-occurrence. Whether it's really a cause, you only find out by testing it.
• Language patterns. How do you write? Recurring words (“have to”, “should”, “tired”), how often “always” and “never” appear, whether you write about a relationship as “we” or as “he” or “she”. Cognitive therapy treats “always/never” statements as a sign of overgeneralisation, a common thinking distortion. And the word-use analyses of Pennebaker and colleagues showed that small, unnoticed words — pronouns, “because”, “I realised” — reveal a lot about how you are processing something.
• Relationship patterns. With whom? Who leaves you energised and who leaves you drained; whose name turns up in your hardest entries; the same conflict with different people. This is the hardest type to see alone, because with each person “it's a different situation”.
• Habit drop-off. When and how do you quit? Most habits don't die in a day; they follow a route: first a weekend is skipped, then “I'll restart on Monday”, then silence. If something fell apart the same way three times in ninety days, that's not a character flaw — it's a recognisable path.

Two of the five — time patterns and drop-off — practically fall out of the dates. The other three are where an AI genuinely sees more than you: it can lay many entries side by side and never gets tired of looking for the same word in the twentieth one.

What a pattern looks like up close

Good pattern recognition is not a label (“you're an anxious type”). It's a chain with five links:

• Trigger: uncertainty, no feedback.
• Thought: “something's wrong”, “they must be angry with me”.
• Behaviour: withdrawing, going quiet, procrastinating.
• Short-term payoff: safety — no need to take a risk.
• Long-term cost: distance, the conversation that never happened, the missed opportunity.

You can see the difference between a generic reply and real pattern recognition on a single entry. Say you write: “I felt anxious when he didn't reply for six hours.” The generic reply: “It's understandable that you felt anxious. Try communicating your feelings.” True, and it says nothing about you.

A pattern-based reply sounds more like this: “Over the past few weeks you've described several similar situations: when there's no clear feedback, you quickly read it as insecurity and pull away before you have more information. Is that what happened this time?” It contains three things the first one lacks: it refers to your earlier entries, it describes the chain, and it ends with a question, not a verdict.

That last part isn't politeness. A pattern is only a hypothesis until you confirm it.

What it takes: consistency, dates, short entries

Good news: pattern recognition doesn't require long or beautiful writing. It requires three things.

• Consistency. Five entries a month is too few; twenty starts to show something. Length matters less than having data from bad and neutral days too, not only the dramatic ones. If you only write when something's wrong, your journal becomes a crisis log — and the pattern gets skewed.
• Date and time. Time patterns come from nothing else. On paper, write down the time too.
• Short but specific entries. Three lines are enough: what happened, what you felt (in one precise word), and a number from 1 to 10. A detail like “after the weekly meeting” is worth more than a page of general mood description.

How long before something shows? The first week is getting used to it and shows nothing. The first repetitions usually start to appear after three to four weeks. Ninety days is a good horizon because by then a pattern has shown up several times, in different situations, and you can tell which ones were coincidence and which ones last. If you want to add a mood score to your journal, our mood tracking guide explains how to do it well.

The 90-day plan — how to do it

You don't need more than ten minutes a day. What matters is when you read back, and what you ask.

1. Days 1–7: getting used to it. Every evening, three lines and a mood score. No analysis, no rereading. The only goal is to tie the entry to something you already do daily: brushing your teeth, your evening tea, the bus home.
2. Weeks 2–4: tags. Keep the three lines, but add one or two tags at the end: who you were with, where, how many hours you slept, whether you moved. These become the raw material for co-occurrence and relationship patterns later.
3. Day 30: the first read. Read the month back in one go — or ask the AI to lay it out side by side. Ask one question: what repeats that I haven't noticed? Write down at most three candidates. Don't decide anything about them yet; just note them.
4. Days 31–60: watching. Keep writing the same way, but now you know what you're looking for. When a candidate shows up again, mark the entry — a star will do.
5. Day 60: filtering. Which of the three candidates came back, and in a different situation too? Let go of anything that happened only once. Whatever repeated becomes your hypothesis.
6. Days 61–75: the two-week test. Change one small thing about one pattern (see the next section) and watch whether anything moves.
7. Day 90: the quarterly picture. Lay the three months side by side. The question isn't how many patterns you found, but which ones share a root. Often two or three patterns that look different grow from the same belief: “I mustn't be a burden”, “if I'm not useful, I don't matter”.

What to do with a pattern: a hypothesis, not a verdict

The biggest mistake you can make with a pattern is to turn it into a verdict: “so that's what I'm like”. It doesn't make you see more clearly; it just makes you feel worse. A pattern is a guess about what's happening — and guesses can be tested.

Two weeks is enough. Here's how:

• Phrase it as if–then. Not “I sleep badly and I'm irritable”, but: “if I sleep less than six hours, the next afternoon I'm more impatient with the kids”. Gollwitzer's research on if–then plans suggests that an intention tied to a specific situation tends to work better than a general resolution — and the same goes for a hypothesis.
• Pick one small change. One, not five: phone down half an hour earlier; a ten-minute walk after the weekly meeting; half an hour on Sunday afternoon to look over the coming week.
• Measure it the way you already do. Same three lines, same number. Don't introduce a new system for the test.
• Compare after two weeks. Did the number move? Did “always” show up less? If yes, the pattern is probably real and you've found a handle on it. If not, that's a result too: either the pattern was wrong or the change was. Neither is a failure on your part.

A made-up but typical example. After three months, someone notices that their tensest entries are written on Thursdays. They suspect work. The two weeks reveal it isn't work at all: on Thursdays they skip their workout to catch up on Wednesday's backlog. They put the workout back, and Thursdays calm down. The pattern was real — the cause just wasn't what they first assumed. That's why finding a pattern isn't enough: you have to test it.

Where the AI gets it wrong — and why you should ask back

An AI is very good at finding repetition. That is exactly why it fails in characteristic ways. They're worth knowing, because a confidently worded but wrong pattern is worse than none.

• It sees patterns in noise. Look at enough data and something will always “correlate”. Humans do this, and so do AIs. Two coincidences are not a pattern.
• It only knows what you wrote. If you write about hard days and not good ones, the picture will be darker than reality. The AI works from your self-report — valuable, but not an objective measurement.
• It may not see everything. AI analyses often receive only a slice of your entries, not all ninety. Whatever was left out, it knows nothing about.
• It tells co-occurrence as cause. “You're irritable after poor sleep” — both might be caused by a third thing, like a stressful project.
• It tends to agree. Language models often confirm what you say rather than contradict it. Hint at an explanation and you'll likely get it back, just more nicely worded.

So the most important skill isn't writing good entries — it's asking good follow-up questions. Four to put to every pattern:

1. “Which of my entries makes you think that?” Ask for a quote or a date. If there's nothing concrete behind it, be suspicious.
2. “How often did it happen, and how often didn't it?” The exceptions tell you as much as the cases.
3. “What else could explain it?” Ask for at least two alternative explanations and see which fits your entries better.
4. “What should I watch for over the next two weeks to find out if it's true?” This question turns a statement into a test.

We've broken down when an AI journal helps and when it doesn't in does AI journaling work. And if you're stuck on what to ask, questions to ask an AI mentor has ready-made ones.

Where the limit is

Pattern recognition is a self-reflection tool. Sometimes you need more than that.

• If a pattern shows lasting, deep distress — you haven't slept or eaten properly for weeks, nothing brings joy — that's not a job for a two-week test. It's a question for a doctor or psychologist. Your journal can help you describe precisely what's happening.
• If a pattern leads back to trauma or abuse, don't unpack it alone with an app. With a professional, though, your journal can be valuable material.
• If searching for patterns becomes rumination itself — analysing every day what's wrong with you — take a break. The goal is a clearer picture, not constant self-scrutiny.
• If an AI observation shakes you, put it down and talk it through with someone. A machine's sentence is not a diagnosis.

And separately: if you find yourself not wanting to live, or wanting to hurt yourself, don't open your journal — get help right now: call your local emergency number (112 in the EU, 911 in the US, 999 in the UK), a crisis line in your country, or a person you can call. That comes before everything else.

Where a system helps

The 90-day plan tends to stall in two places. One is scattering: the mood score lives in one app, the journal in a notebook, sleep somewhere else — and on day 30 there's nothing to lay side by side. The other is rereading: going through thirty entries and comparing them is tiring, so most people skip it.

In Mirrify, your journal, morning and evening reflections, voice journal, mood tracking, habits and goals all collect in one place, with dates. On the Pro plan, the AI mentor's deep analyses include pattern analysis, and it also produces weekly, monthly, quarterly and yearly reviews. The deeper analyses quote your own words so you can check what they're based on — which answers the first follow-up question above before you even ask it. The AI journal search lets you find when you've written something similar before. How the mentor builds on your earlier entries is covered in the AI mentor that remembers.

Two things are worth knowing up front. First, an analysis doesn't read your entire journal. Currently at most 30 entries go into one, each up to roughly 800 characters — which is why short, dated entries work well, and why it pays to ask period by period. The technical details are in what the AI sees in your journal. Second, the Basic plan includes no AI analysis; Pro can be tried free for 14 days. That's enough to settle in and start tagging, not for the full ninety days — better to know that in advance.

What it doesn't do: it doesn't diagnose, it isn't therapy, and it doesn't decide for you what's true about you. The AI suggests the pattern — your two weeks decide whether it holds.

The one sentence worth taking away

If you take away one thing, make it this: a pattern isn't a verdict about you, it's a guess you can test in two weeks — but first you have to write down the ordinary days too, with dates.

And if you do just one thing now: tonight, write three lines and a number from 1 to 10, and note the time next to it. That's day one of ninety.

Frequently asked questions

How does AI recognise patterns in my journal?

It lays several entries side by side and looks for what repeats: the time of day, the situation beforehand, the words you use, the people involved. That's hard for a person, because nobody holds thirty entries in their head. But the AI only works from what you wrote, and often sees only a slice of your journal, so its observation is a hypothesis, not a fact.

How long does it take for recurring patterns to show?

The first week is just getting used to it; nothing shows yet. The first repetitions usually appear after three to four weeks of consistent writing. Over ninety days, a pattern shows up several times in different situations, so you can tell coincidence from something lasting.

What kinds of patterns can a journal reveal?

Five main types: time patterns (time of day, day of the week), co-occurrence patterns (for example, less patience after poor sleep), language patterns (recurring words, “always” and “never”), relationship patterns (who leaves you drained or energised), and habit drop-off patterns (how and when you quit things).

How much do I need to write for the AI to find patterns?

Less than you think. Three lines a day is enough: what happened, what you felt in one precise word, and a number from 1 to 10. Consistency and dates matter more than length. Write about neutral and good days too, not just hard ones, or the picture will be skewed.

Can the AI be wrong about my patterns?

Yes, in typical ways: it sees patterns in coincidence, works only from what you wrote, presents co-occurrence as cause, and tends to agree with you. So ask back: which entries is it based on, how often did it happen and not happen, what else could explain it, and what should you watch for two weeks to find out.

What should I do when the AI finds a pattern?

Treat it as a hypothesis, not a verdict. Phrase it as if–then, choose one small change, and measure it for two weeks with the same three lines and number as before. If something moves, the pattern is probably real. If not, that's a result too: either the pattern or the change was off.

Can an AI diagnose me from my journal?

No, and it shouldn't be used that way. Pattern recognition is a self-reflection tool: it shows what repeats in your writing, but it doesn't identify illness. If you see lasting distress in your journal, talk to a doctor or psychologist. In a crisis, call your local emergency number or a crisis line.