When to Stop Running a Social Connection Activity

Oregon State University published a finding in May that should unsettle anyone running a connection programme on the assumption that contact is contact. Among adults aged 30 to 70, connecting with strangers online was linked to greater loneliness. Connecting online with people already known wasn’t linked to more loneliness — but it wasn’t linked to less, either. Being in touch is not the same as being less alone.

That distinction matters well beyond social media. It is the exact question a social prescribing lead should be asking about the walking group that’s run for two years, the befriending scheme with a loyal roster, or the online interest forum a commissioner funded because it was cheap to scale. Attendance is contact. It is not automatically outcome.

The attendance trap

Call it the attendance trap: a programme survives review after review because the room is full, not because loneliness scores have moved. Full rooms are easy to see and easy to defend to a funder. Loneliness reduction is harder to see and requires you to have measured it in the first place. Programmes that produce the first kind of evidence without the second tend not to get stopped — they get renewed, because nobody wants to be the one who closes something people show up to.

The Oregon State result is a clean illustration of why that’s a mistake. The strangers-online cohort in that study was presumably engaged — logging in, posting, replying. Engagement was not the problem. The absence of a loneliness benefit was. A social prescribing group with excellent attendance can be in exactly the same position, and you won’t know unless you’re tracking the outcome rather than the proxy.

The parallel research on AI companionship makes the same point from a different angle. Reporting on companion chatbots has consistently found short-term comfort alongside longer-term drift toward more loneliness with heavy use — a pattern that shows up only if someone measures at more than one time point. A single satisfaction snapshot after a session would have missed it entirely.

Three questions before you renew anything

1. Do you have a loneliness or isolation measure at more than one point in time? Attendance, satisfaction, and “participants said they enjoyed it” are not outcome data. If the only thing you can report is that people came back, you cannot yet answer whether the activity is working.

2. Is the activity structured around a specific mechanism, or is it just co-presence? The qualitative literature on social prescribing has consistently found that participants describe benefit coming from restored purpose and meaningful participation, not contact for its own sake. A structured, purposeful activity has a theory of why it should reduce loneliness. Unstructured co-presence — a shared room, a shared feed — often doesn’t, and the Oregon State findings on passive online contact are a data point for exactly that gap.

3. Have you compared it to a plausible alternative, not just to nothing? Two randomised trials in the loneliness literature that are unusually direct on this point: one found volunteering-based prosocial engagement reduced loneliness among older adults; another, more recent, found telephone-delivered behavioural activation and mindfulness outperformed a befriending comparison group at twelve months in a much larger trial of older adults living in poverty and digitally excluded. Befriending — the single most common social prescribing intervention — was the losing arm. That doesn’t mean befriending fails. It means “people like it and it feels supportive” is not sufficient grounds to assume it’s the best use of a caseload slot.

What this means in practice: before you renew a group, activity, or pathway, ask whether you can point to a loneliness measure that moved, not just a register that filled. If you can’t, the honest answer is that you don’t know if it’s working — and “we don’t know” is a legitimate reason to pilot something with measurement attached, rather than to keep funding it on faith.

Evidence status, by intervention type

Claim	Evidence status
Structured group activity with a clear purpose reduces loneliness	Reasonably supported — repeatedly identified as the mechanism in qualitative synthesis
Befriending reduces loneliness	Supported, but outperformed by structured psychological approaches head-to-head
Volunteering/prosocial engagement reduces loneliness in older adults	Supported by one of the few RCTs in this field
Online contact with strangers reduces loneliness	Not supported — new evidence links it to more loneliness
Online contact with people already known reduces loneliness	Not supported either way — no measurable benefit found
Social prescribing overall reduces loneliness	Encouraging but thin — systematic review protocols still note only one peer-reviewed RCT for older adults specifically

What this doesn’t solve

None of this tells you how to close a group that people value emotionally but that isn’t moving a validated measure — that’s a staffing and communication problem, not a measurement one, and it’s often the harder conversation. It also assumes you have the capacity to measure loneliness longitudinally in the first place, which many social prescribing services running on short-term grants do not. And it says nothing about reach: a rigorous stop/continue process only ever evaluates the people who showed up. It has nothing to say about the people who were referred and never came, or who were never referred at all.