How to Figure Out What Actually Brings You Joy (Not What You Assume Does)

By Matt Hughes · 2026-08-03 · 7 min read

Kansha's AI reads back through your entries and scores your five biggest sources of joy from 0-100 — deliberately spread out, not bunched near the top. Here's why that specificity matters more than it looks.

Ask most people what actually brings them joy and you'll get an answer that's technically true and completely useless: family, friends, weekends, maybe the dog. It's not wrong. It's just too broad to do anything with. You can't act on "family" the way you can act on "the twenty minutes on Tuesday when my brother sent a voice note that made me laugh on the train." One of those is a category. The other is a source.

That gap — between the vague answer we give and the specific thing that's actually doing the work — is what I built Kansha's "themes" feature to close. It's one of the quieter parts of the app's AI insights, and it's had zero coverage on this blog until now, which is a bit of an oversight given it's the part I find myself checking most.

Why "family and friends" isn't a useful answer

There's a strand of psychology research on what's called emotional granularity — how precisely you can name what you're feeling. Todd Kashdan, Lisa Feldman Barrett and Patrick McKnight's 2015 review in Current Directions in Psychological Science pulls together the evidence: people who can tell "frustrated" apart from "disappointed," or "content" apart from "proud," regulate their emotions better. They're less likely to reach for a drink or lash out when things go wrong. Low granularity — lumping everything into "bad" or "good" — shows up more often in anxiety, depression, and a handful of other conditions where people struggle to manage what they're feeling.

Almost all of that research is about naming difficult emotions with more precision. But the same logic runs the other way. If "I feel bad" is a useless starting point for figuring out what to do about it, "I feel grateful" or "today was good" is just as useless for figuring out what to protect or repeat. Specificity is the thing that turns a feeling into information. That's the whole argument for this feature: not a nicer way of saying "grateful," but a more precise one.

What Kansha's AI actually does with your entries

Here's the mechanism, no mystery to it. When you've written enough entries, the same AI that generates Kansha's insight cards also reads back through your journal and produces exactly five objects — the five biggest categories of what brings you joy, based on what you've actually written, not a personality quiz you filled in once. Each one gets a score from 0 to 100 for how prominently it shows up across your entries.

The instruction I gave the model for that scoring is deliberately blunt: be honest, don't bunch everything near 90. That line exists because the first version of this feature I tested gave nearly everyone a wall of scores in the high 80s and 90s — technically positive, completely useless, exactly the "family and friends" problem in numeric form. A score only means something if it can go down as well as up, and if the five scores are actually spread out rather than clustered.

This is a different feature from Kansha's general pattern insights — the cards that might tell you "you mention sleep quality more on weekdays" or flag someone who's gone quiet in your entries. Those look for patterns across everything you write. Themes narrows in on one question: out of everything that's shown up as a source of joy, what are the five biggest, and how do they rank against each other. It's the joy-specific slice of a broader analysis engine.

Why five, and why that number matters more than it sounds

Five is a design choice, not a number handed down by a study — I picked it because it's few enough to read at a glance and enough to force some actual differentiation. But the instinct behind wanting several categories rather than one dominant score has real research behind it, from a different angle to the granularity work above.

Jordi Quoidbach and colleagues published a study in 2014, in the Journal of Experimental Psychology: General, using survey data from more than 37,000 people. They measured what they called emodiversity — the variety and balance of the emotions someone experiences, not just how positive or negative they feel on average. People with higher emodiversity had better mental and physical health outcomes — fewer depressive symptoms, fewer doctor's visits — and this held up even after accounting for how much positive or negative emotion they reported overall. The finding wasn't "feel more good things." It was "feel a wider spread of things," and that spread mattered independently of the average.

Translate that to joy specifically, and the implication is that someone whose entire sense of what's good in their life rests on one theme scoring 95 is in a more fragile position than someone with five themes spread across 40 to 75, even if the second person's average score is lower. One pillar is a liability. Several, even modest ones, is closer to resilience. That's the quiet argument underneath a feature that, on the surface, just looks like a little scored list.

What the scores are actually good for

The obvious use is validation — seeing "time outdoors" or "small wins at work" land near the top and thinking, yes, that tracks. That's fine, but it's the least interesting use of the feature. Three things I'd actually do with it:

Look at the theme with the lowest score, not the highest. If something is on the list at all, the AI found it recurring enough across your entries to count as one of your five biggest joy sources — and it's still your lowest. That's not a category to write off. It's usually the one that's true but under-fed: something that reliably makes you feel good when it happens, but doesn't happen often enough to score higher. Worth deliberately doing more of, rather than waiting for it to show up on its own.

Watch how the five change over months, not weeks. A theme dropping isn't necessarily bad news — it might mean a phase of your life wound down naturally. But if something that used to be near the top quietly disappears from the list altogether, that's worth a second look, in the same way Kansha's separate "went quiet" insight flags people or activities that have stopped appearing in your entries.

Notice when four out of five themes sit close together and one is way out ahead. That's the "everything near 90" problem showing up in a more honest form — a genuine concentration risk, per the emodiversity research above, not a flaw in the scoring.

Where this falls short

I want to be straight about the limits, because it would be easy to oversell a feature I built. This is a language model reading text and making an inference, not a clinical instrument. If you write two lines a day, tersely, the themes it extracts will be thinner and noisier than if you write properly. If you go through a stretch of writing almost nothing, the five themes it returns will lean on older entries and may not reflect where you actually are right now. And the categories themselves are the model's phrasing of what it read, not a fixed taxonomy — ask it again after another month of entries and the wording might shift even if the underlying pattern hasn't.

None of that makes it useless. It makes it a mirror with a bit of noise in the glass, which is roughly what any tool built on top of your own free-text writing is going to be. The honest version of the pitch isn't "this tells you what brings you joy." It's "this takes something you already know in a fuzzy way and forces it into five specific, rankable pieces" — and specificity, per Kashdan, Barrett and McKnight, is most of the value.

The bigger point

I didn't build this to produce a pie chart. I built it because "what brings you joy" is a question almost everyone answers too generally to use, and a journal — genuinely read back, not just written into — is one of the only places with enough raw material to answer it properly. The themes feature is Kansha trying to do, mechanically, what the granularity research says is worth doing anyway: naming the specific thing, not the category it belongs to. Five scored, ranked, occasionally uncomfortable specifics beat one comfortable generalisation every time.