For my research I spent a long time on a question that turns out to be more revealing than it first sounds: what do employees say about their work on public social media, and what does it tell us? The project ended up under the title I Tweet About My Work, and it won a research prize, which was lovely — but the reason I want to write about it here is the finding underneath, because it's a small, clear example of the thing I keep circling: that data is people communicating, and reading it well means taking the communication seriously.

The premise: people talk, publicly, all the time

Organisations spend enormous effort on their official voice — the careful corporate account, the approved messaging. Meanwhile, their employees are out there on public social media, in their own voices, mentioning their work constantly: a frustrated line after a hard day, a genuinely proud post about a project, an offhand joke about a process, an enthusiastic share of something the company did. Individually, each is trivial. In aggregate, they're a large, continuous, largely unguarded stream of signal about what it's actually like inside that organisation.

The research question was really: if you gather and read that stream systematically, what can you learn? And the answer was: quite a lot — often things the organisation's official channels would never tell you, and sometimes things the organisation doesn't know about itself.

What the aggregate reveals

Reading many employees' public work-talk together, patterns emerge that no single post contains. You start to see the genuine mood beneath the official one — whether the pride is real or performed, where the recurring frustrations cluster, which values the company preaches that its people actually echo and which land with silence. Employees, speaking casually in their own voices, collectively paint a picture of the organisation's real culture that is often strikingly more honest than anything in its formal communications.

That's the finding that's stayed with me. An organisation has two voices: the one it broadcasts deliberately, and the one that emerges from its people talking unguardedly. The second is messier, harder to read, and considerably more truthful — and it's sitting in public, if you know how to gather and interpret it.

The method, and its honesty problem

Doing this properly meant everything I've written about content analysis before: collecting the posts, and then the hard, human work of coding them — deciding what counts as positive or critical, what themes each touches — with careful rules and checks so the judgements were consistent rather than just my own impressions. The temptation with social-media data is always to let a tool declare the sentiment and call it objective. But sarcasm, in-jokes, and context defeat tools constantly, and a workplace grumble can read as praise to a dictionary that only matches words. So the interpretation stayed human where it needed to, and automated only where the rules were genuinely reliable.

There's also a real ethical edge I won't gloss over: this is public data, but it's people's voices, and treating a person's offhand post as a data point carries a responsibility to handle it with care and in aggregate, not to expose individuals. Reading the collective signal is fair game; turning it on a single named employee is a different and murkier thing.

A pattern that surprised me

One finding has stayed with me because no official channel would ever have surfaced it. Reading the posts in aggregate, the genuine pride clustered tightly — it gathered around specific teams and specific kinds of work, while going conspicuously quiet elsewhere. And the low-level frustration clustered too, recurring around a particular process that the same people mentioned, in passing, again and again.

Neither of those was anywhere in the organisation's formal communications, which spoke in a single, even, on-message voice. But the employees, posting casually and separately, mapped out — without meaning to — where the real energy was and where the real friction sat. No one set out to reveal it; it emerged from a thousand small, unguarded posts read together.

That's the thing I keep coming back to: the signal wasn't in any single tweet, which on its own was trivial. It was in the aggregate — and it was truer than the official story precisely because nobody was managing it.

Why it matters beyond the prize

Here's why this little project sits at the root of how I think. It's the clearest demonstration I've found of a single idea: that data is, very often, people communicating — and that if you read it as mere numbers, you miss most of what it's actually saying.

A naïve analysis of this data would count positive and negative posts and produce a tidy sentiment percentage. But the meaning is in reading it as communication — understanding that an employee's post is a person saying something, to someone, for a reason, in a context — and the richest findings came precisely from taking that seriously rather than flattening it into a score.

Employees tweeting about their work don't set out to reveal their organisation's real culture. They do it anyway, one unguarded post at a time — and read together, honestly, it's one of the truest pictures you can get.

I came out of this research more convinced than ever that the interesting frontier isn't better tools for counting words. It's the discipline of reading data as what it usually is underneath — human communication — and refusing to let the convenience of a number erase the meaning that made the data worth gathering in the first place.