The Vanity Metrics Trap: Why Citizen Science Does not Need to Go Viral

I have a confession to make, and it is not a particularly glamorous one: for a long time, I judged the health of a citizen science project the same way I judged everything else on the internet, by counting hearts.
A post about a bird-monitoring campaign gets three thousand likes and I feel a small surge of professional satisfaction, as if those three thousand little red icons were somehow equivalent to three thousand data points sitting neatly in a spreadsheet. They are not. They never were. And the longer I have spent talking to researchers who actually run these projects, the more I have come to suspect that this confusion between attention and participation is not a minor communication quirk.
It is, quite possibly, the single most damaging assumption in the entire field.

The Like That Isn’t a Sample

Here is the uncomfortable arithmetic. A viral post is a broadcast event: one voice, reaching many ears, most of which will forget the message within the hour. A citizen science project, on the other hand, only exists insofar as people do something afterward, repeatedly, correctly, and over time.
If nobody photographs the hailstones, counts the frog calls, or measures the rainfall in their backyard, then what you have built is not a research infrastructure. It is entertainment with a scientific costume on.
I say this with genuine affection for entertainment, which has its own dignity, but a like is not a sample, and confusing the two is how projects end up with beautiful engagement graphs and unusable datasets.

From Broadcast to Horizontal

The good news is that somewhere between the death of the mass Instagram campaign and the rise of something quieter, a different model has been taking shape, and I think it deserves far more attention from research teams than it currently gets.
Call it the migration from broadcast to horizontal: instead of one institutional account shouting into a feed and hoping for the best, projects are building small, closed, almost stubbornly focused digital spaces, a Discord server with dedicated channels, a Telegram group for fast alerts, a subreddit that nobody outside the niche has ever heard of.
Take INTENT, an ERC-funded project on AI-driven musical instruments based at the Iceland University of the Arts: alongside its open-source releases, the team runs regular workshops and maintains an active Discord server where the public and university students take part directly in the thinking behind the research, not just its consumption. It is a small infrastructural choice, but it changes the entire social contract of the project: from an audience being informed to a group of collaborators actually working together.

Why Fifty Trained People Beat Five Thousand Strangers

I want to be careful here, because I know how this sounds, and I do not want to romanticize smallness for its own sake. Fifty people are not automatically better than fifty thousand. But there is real, unglamorous evidence that a smaller, better-trained group tends to produce cleaner data than an enormous, loosely engaged crowd, and the mechanism is not mysterious once you look at it.
In one study on crowdsourced frog call identification, ordinary volunteers correctly identified species only a little more than half the time, and that accuracy only climbed meaningfully once researchers required agreement between at least two out of five separate volunteers. In other words, scale did not fix the noise, redundancy and training did.
A working paper produced after the second Austrian Citizen Science Conference makes a related point rather bluntly: the variability introduced by differences in volunteer knowledge, skill, and motivation, along with the temptation to oversimplify tasks to attract more people, is itself one of the main sources of poor data quality. Fifty motivated, trained people in a Discord server, checking each other’s photographs and arguing gently about edge cases, will very often out-produce five thousand strangers who clicked through once and never came back.

The Speed Advantage of the Small Room

There is also a second, more practical advantage to the small room, one that anyone who has ever tried to organize anything in a group chat versus a mailing list will instantly recognize: speed. Some of the most useful citizen science data is only valuable if it arrives within minutes, not days.
NOAA’s mPING app was built precisely for this kind of urgency, asking ordinary people to report hail or snow the moment it is falling on their roof or driveway, rather than after the fact, and the CoCoRaHS network built an entire hail-reporting pipeline, now feeding NASA’s research on how hailstones melt as they fall, around the same logic of near-instant, on-the-ground observation submitted the moment a storm occurs so the reports are immediately available for use.
You do not need twenty thousand volunteers to replicate this effect at a smaller scale. You need a group small enough that a single message, “it’s hailing right now, get your rulers out,” actually lands in front of people who are paying attention and physically nearby. A public feed cannot do that. A focused chat can, almost embarrassingly well.

The Trust Dividend

None of this works, though, without the third ingredient, which is less about data pipelines and more about something closer to friendship, or at least professional intimacy. Researchers who have written honestly about sustaining volunteer communities keep circling back to the same lesson:

people stay engaged not because the science is compelling in the abstract, but because someone with a research badge actually shows up and talks to them.

The team behind the citizen science project Redshift Wrangler found that once they introduced direct touchpoints, welcome videos from the research team and regular office hours, volunteers began participating more actively in discussions and sharing more about their own motivations and experiences, rather than only submitting classifications and disappearing.
That is the trust dividend of the small room: it is very hard to ghost a project when the person who explained the methodology last week is the same person who will notice, kindly, that you have been quiet.

Building the Room: A Practical Path

So if you are persuaded, at least provisionally, that fifty engaged people might be worth more to your research than fifty thousand passive ones, the practical path is not especially mysterious, though it does require resisting the urge to open the doors to everyone.
Start by choosing a platform that matches the texture of your work rather than its marketing potential: Discord tends to suit projects that need organized, topic-based channels and a slower, browsable archive, while Telegram often wins when speed and mobile immediacy matter more than structure, as in that hailstorm scenario.
Resist, too, the instinct to let anyone in automatically. A short onboarding form, nothing bureaucratic, just enough to ask why someone wants to join and what they hope to contribute, does more to filter for genuine motivation than any algorithm ever will, and it sets the tone that this is a working group, not a fan page. And once people are in, feed them back their own impact relentlessly: show them the map that changed because of yesterday’s reports, name the pattern their observations revealed, make the invisible loop between contribution and discovery visible on a weekly basis.
Gamification badges are fine as decoration, but the real reward, the one that keeps a fifty-person Discord server alive for years, is watching your own small act of noticing become part of something a scientist actually needed.

I will admit that “our project reaches ten thousand people a month” sounds better in a grant report than “our project has a stable core of fifty trained contributors,” and I am not naive about the institutional pressures that push communicators toward the first sentence. But I would rather defend the second one in front of any funding committee, because it is the one that is actually true to what citizen science is supposed to be: not an audience, but a collaboration.
So here is the question I keep asking myself, and the one I would genuinely like to ask you. Would you rather have ten thousand sterile views, or fifty perfect, usable, scientifically actionable reports? And more to the point, which one is your project actually optimizing for right now?