Why do I keep losing friends as an autistic adult?

What the research finds

Wu and Wang’s 2025 scoping review in Frontiers in Psychiatry is the only review of its kind, and it found 22 empirical studies in total. That number is worth holding on to. This is a thin literature carrying a lot of weight.

Pooling four studies that used the Friendship Questionnaire, it reported a mean of 61.82 for 782 autistic participants against 76.69 for 924 non-autistic participants. Two studies using parent reports found that 60.2% and 88.3% of autistic adults had at least one close friend—a spread wide enough that the review attributes it to who was answering and what counted as a friend.

The closeness measures disagree with each other. Pooled across four studies, autistic participants scored 5.21 on the Unidimensional Relationship Closeness Scale against 6.13 for non-autistic participants. But Sedgewick, Leppanen and Tchanturia’s 2019 study of 949 adults found the opposite on that same scale: asked about one specific close relationship rather than friendships in general, autistic participants rated it more highly than non-autistic adults did. One study going the other way isn’t a finding, but it is a caution against reading the averages as a verdict on your closest friendship.

The difference is in the practice

This is the review’s own conclusion rather than an interpretation laid over it.

Across two studies with 228 autistic and 232 non-autistic participants, autistic adults preferred lower levels of closeness, both emotional and physical. They mostly saw friendships as opportunities for enjoyment rather than as sources of support. Non-autistic participants reported the reverse—their friends valued them as someone who supports them.

The specifics get granular. Autistic young adults preferred to phone a friend to make arrangements and to meet up for a particular activity. Non-autistic participants preferred to phone a friend to chat, and to meet up to chat. Contact patterns split the same way: in-person visits with a close friend happened every few weeks for 27.7% and less than monthly for 24.6%, while electronic contact ran several times a day for 29.2%.

The review also looked at how friendship gets defined in the first place. Comparing autistic and non-autistic adults across six components of friendship understanding, one study inside it found differences in both the motivational parts—intimacy, support, companionship—and the developmental ones, including reciprocity and complexity. Ten of the review’s studies were qualitative, and in those, people described friendship in their own words as shared interests, humour, and benefit, with autism not necessarily central to it at all.

Read that as two working definitions of the same word. One person thinks friendship means frequent unstructured talking and emotional support on call. The other thinks it means doing a specific thing together, with a lot of text in between. Neither is doing it wrong. But if the first person is measuring the friendship by the second person’s phone habits, they’ll conclude they’re being dropped—and the second person, who thought things were fine, finds out only when it’s over.

The mismatch account, and how it’s holding up

Damian Milton’s 2012 “double empathy problem” argued that the breakdown between autistic and non-autistic people is mutual rather than a deficiency in the autistic person. It’s a theoretical paper in Disability & Society, and it reframed a great deal of subsequent research.

The evidence for it is mixed, and anyone telling you it’s settled is overselling.

Where it holds up. Morrison and colleagues in 2020 put 67 autistic and 58 non-autistic adults into five-minute conversations with strangers, in three kinds of pairing. Autistic participants were rated as more awkward and less socially warm by both autistic and non-autistic partners. But only the non-autistic participants preferred future interaction with non-autistic partners over autistic ones. Autistic participants trended toward preferring other autistic people and reported disclosing more to them. The authors read this as social difficulty being relational rather than located in one person.

Where it didn’t survive testing. Crompton and colleagues in 2020 passed a story along chains of eight people and found that information degraded in mixed autistic and non-autistic chains but not in single-neurotype ones. That study had 72 participants and became the headline empirical support for double empathy. In 2025 the same lead author ran it again as a registered report with 311 people and found no difference in information transfer between mixed and single-neurotype chains. Wilks and colleagues ran a parallel study with 310 adults on a visuospatial task and found no mixed-chain breakdown either. Non-autistic chains did report higher rapport, and disclosing diagnosis improved rapport.

So the rapport and preference findings are standing up better than the information-transfer ones. Reporting that accurately matters more than defending a theory: the broader mismatch account remains plausible, but the specific mixed-chain information-transfer effect did not replicate. The 2020 study reported selective degradation across mixed chains, not an inability to communicate, and it’s that narrower effect that later work failed to reproduce.

What other people bring to it

There’s a strand of research that puts the variable somewhere people rarely look for it.

Sasson and Morrison had 215 observers rate 40 adults from brief videos in 2019. First impressions of autistic adults were less favourable—and improved when observers were accurately told the person was autistic. In a companion study, 505 raters judged 20 autistic adults, and variability in the ratings was driven more by characteristics of the raters than of the autistic adults. Rater stigma and autism knowledge did the work. Only ratings of “awkwardness” tracked the autistic adults more than the raters.

One finding there deserves care rather than enthusiasm: telling people made impressions better on average, and made them worse among raters who already held high stigma toward autism. Disclosure isn’t a reliable repair, and who you’re telling changes what it does.

What tends to help

No study has compared approaches to keeping friendships, so what follows is reasoning from the findings above rather than a tested recipe.

Name your cadence instead of leaving it to be read. If you go quiet for three weeks and that means nothing is wrong, the research suggests your friend may not read it that way. Saying so once costs less than the misreading does.

Look for shared-interest settings. In one study of 20 autistic adults inside the review, the people who built satisfying friendships did it with people who accepted their social differences, and found them through shared interests rather than through general socialising.

Notice which friendships ask you to perform. The review found a gap between the friendship people wanted and the friendship they had. A friendship that only works while you’re masking is one of the ways that gap opens.

Autistic company is a reasonable thing to seek out. The interaction-preference data is one of the more robust findings here, and it’s the one people act on least. The review’s own recommendation lands in the same place: autistic-led social opportunities, and settings built around a shared focus rather than around open-ended socialising.

It’s worth separating two things that get bundled together. The evidence that autistic adults have fewer friendships on average is reasonably consistent. The evidence about why is thin, contested in places, and has never been applied to an individual case.

If you’ve read this far looking for the flaw that explains the losses, the research doesn’t supply one. It supplies two groups with different defaults, and a lot of evidence that the misreading runs both ways.