Matt Klein, Head of Foresight at Reddit and founder of ZINE, spends months each year manually sifting through global trend reports to get a feel for how culture is actually moving. Reading more than 70 forecasts side by side, he looks for the patterns, tensions and contradictions that machines tend to smooth over. In conversation with Protein XYZ editor John Sunyer, Klein reflects on what emerges when you slow down and read culture by hand – and why intuition, ambiguity and lived experience still matter in an age of dashboards, metrics and AI.

Protein This year's META Report feels unusually deep and intentional – almost archaeological in how it’s put together. Can you outline the process a bit? How long does something like this actually take? How many people are involved? 

Matt Klein It’s just me, however each year I have a friend lend a mind to ensure I’m not off base. It’s roughly a three-month process and it usually starts around mid-November, when all the annual reports begin to drop. From there, it’s about gathering everything and reading through it all.

I’ve definitely gotten faster over time – I know which reports are garbage and which are actually worth digging into. I won’t name names, but anyone who’s been doing this work long enough knows which ones are empty calories.

From there, it’s extremely manual. I parse out individual trends from each report, capture how they’re described, and then do that over and over again. Eventually I start colour-coding: okay, green themes feel connected; these pink ones are circling around something about overstimulation; these others are about slowing down or re-orienting life.

Just by sitting with all of that text, you start seeing connections – this links to that, that echoes this. And the important thing is I don’t want to just dump this into AI and have it spit out themes. I keep trying and it doesn’t work. This is about developing a pulse for what people are actually talking about. You can’t outsource that to a machine, which isn’t experiencing culture.

Culture isn’t coded – it’s lived. Pattern recognition here is a practice. And it’s also longitudinal: how do these ideas mutate year after year? That immersion gives you something no AI can replicate.

Once everything’s clustered, I’ll use AI in a very specific way – pulling representative keywords from each cluster of trend description and scoring them by social listening