Before You Borrow a Loneliness Programme, Check Who It Was Built For

A befriending scheme built for isolated 80-year-olds gets pitched to a workplace wellbeing budget as a fix for lonely 24-year-olds. A volunteering intervention tested on older adults in Hong Kong gets cited as evidence for a youth mentoring programme. A model designed around weekly home visits gets compressed into a Slack channel for remote staff. This happens constantly, and it usually happens because the programme worked somewhere, and “worked somewhere” gets mistaken for “will work here.”

It might. But the two populations rarely have the same problem. They just share a word.

The word “lonely” hides at least three different problems

Loneliness in adults over 65 is disproportionately driven by shrinking networks: bereavement, retirement, reduced mobility, the physical loss of the people and places that used to generate contact. The National Academies’ 2020 framing of isolation as a health-system issue for older adults rests on this: the problem is often structural absence, not skill or motivation.

Loneliness in adults under 30 looks different. The American Enterprise Institute’s 2021 survey found the share of American men reporting no close friends had risen fivefold since 1990, and Harvard’s Making Caring Common project found 61% of 18-to-25-year-olds reporting serious loneliness in 2021, with 43% saying it had gotten worse since the pandemic began. This is not a shrinking network in the way an 80-year-old’s is. Younger adults often have more nominal contacts than older adults do. What they report lacking is depth — someone who would notice if they disappeared, someone who takes more than a few minutes to ask how they’re doing and mean it.

Loneliness among employees is a third thing again. Gallup’s 2024 workplace data found remote employees reporting loneliness at 25%, against 16% for fully on-site staff — a gap that tracks physical proximity, not friendship depth or network size. That is closer to the older-adult mechanism (absence of incidental contact) than to the young-adult one (thin relationships despite plenty of contact).

Three different mechanisms. One word. If your programme design assumes the word, and not the mechanism, you will build something that treats the wrong problem well.

What this looks like when it goes wrong

The reunion problem. A befriending or visiting model built for older adults typically works by re-establishing regular, low-effort, one-to-one contact with a consistent person. Applied to lonely young adults, this can land as pity or surveillance rather than relief — the AEI and Harvard data both suggest younger adults’ loneliness is less about lacking any contact and more about lacking contact that feels reciprocal and unscripted. A scheduled call from an assigned volunteer does not obviously fix that; it can even underline the gap between assigned care and organic friendship.

The volunteering transplant. The HEAL-HOA randomised trial — one of the few controlled trials in this literature, run with lonely older adults in Hong Kong — found volunteering and prosocial engagement reduced loneliness in that group. This gets cited, reasonably, as some of the strongest available evidence that a loneliness intervention actually works. It gets cited less reasonably as grounds for prescribing volunteering to lonely 22-year-olds, on the assumption that “helping others helps the helper” is population-agnostic. It might transfer. But the trial’s population had different baseline network structures, different amounts of unstructured time, and different reasons for isolation than a young adult juggling multiple jobs and a thin friend group. Nothing in the trial tests that transfer.

The third-place mismatch. Ray Oldenburg’s account of the third place and Eric Klinenberg’s account of social infrastructure both describe institutions — cafés, libraries, barbershops, union halls — that generate repeated, low-stakes, unplanned contact over months and years. That mechanism depends on people returning to the same physical space on a recognisable schedule. It works well for retirees and for neighbourhood-anchored populations. It works less well for a population defined by irregular shifts, frequent moves, or remote work, because the precondition — showing up in the same place repeatedly — is exactly what their life structure prevents. Recommending “join a regular class at the library” to someone working rotating warehouse shifts is not adapting the model. It is ignoring why the model worked in the first place.

A short checklist before you adapt anything

1. Name the mechanism, not the outcome. “Reduces loneliness” is an outcome. “Restores predictable weekly contact with a familiar person” is a mechanism. Write down the mechanism the original programme actually relies on.
2. Check whether your population has the precondition the mechanism needs. Regular contact models need people with a stable, repeatable schedule and a fixed location. Depth-of-relationship models need people who already have some contact but little reciprocity. Structural-access models need people who are missing a place to go, not a person to talk to.
3. Check what social prescribing evidence actually supports. Qualitative work on social prescribing consistently finds that participants describe benefit in terms of restored purpose and meaningful participation, not contact for its own sake — structured, purposeful activity outperforms unstructured contact. A 2025 systematic review protocol on social prescribing for older adults notes that despite wide adoption, only one peer-reviewed randomised controlled trial exists for this population. If you are borrowing a social prescribing model wholesale, you are borrowing something with a genuinely thin evidence base, whatever population you apply it to.
4. Pilot with the new population before scaling. Not because piloting is good practice in general, but because the mechanism check in step 2 is a hypothesis, not a guarantee.
5. Measure the mechanism, not just the headline outcome. If you can’t show the precondition held (people did return repeatedly; contact did deepen; purpose was restored), a drop in loneliness score tells you little about whether you can repeat the result with a third population.

What this means in practice: don’t ask “did this programme reduce loneliness in its original population?” Ask “what specific precondition of that population made the mechanism work, and does my population share it?” If the answer is no, you’re not adapting the model. You’re running a different, untested intervention with borrowed marketing.

Evidence status

Claim	Status
Older adults’ loneliness is driven substantially by shrinking networks and reduced mobility	Well supported (National Academies framing, AARP’s comparable 2018/2025 surveys)
Younger adults report loneliness despite larger nominal networks; depth of contact is the gap	Reasonably supported by survey data, not causally tested
Volunteering reduces loneliness in isolated older adults	Supported by one randomised trial — rare strength for this field, but a single population and setting
Volunteering or prosocial engagement reduces loneliness in younger or working-age adults	Untested; extrapolation from the above, not evidence in its own right
Social prescribing works better as structured, purposeful activity than as contact alone	Supported by qualitative synthesis; underlying trial evidence remains thin
Third-place models transfer to populations without stable schedules or fixed locations	No direct evidence either way; theoretically weak given the model’s dependence on repetition

What this does not solve

This checklist tells you how to avoid transplanting a mechanism into a population it wasn’t built for. It does not tell you what to build instead if none of the available evidence base fits your population — and for several groups (shift workers, new parents, people who move frequently), it largely doesn’t. It also does not solve the reach problem common to almost everything in this field: adapted or not, most of these programmes still only reach people who already showed up looking for help, which is a small and self-selecting slice of anyone’s target population.