Friendships don’t end. They drift.

The Utrecht panel study is the clearest measurement of this. Researchers interviewed 1,007 people about their personal networks, then went back to 604 of them seven years later and asked again. Under half the original network was still there. For the closest roles — the people you discuss important matters with, the people who help you out — only around 30% still held the same position.

What makes it hard to notice is that nothing appears to be missing. People replaced the ties they lost, so the total stayed flat. You do not experience your network shrinking. You experience it being roughly the same size, made of different people, and you rarely sit down and compare the list against the one from seven years ago.

That is why "drift" is a better word than "loss". Nothing is dropped. It slides.

Roberts and Dunbar followed people over eighteen months and found an asymmetry that explains a lot of ordinary experience: emotional closeness to friends declined steadily once regular contact stopped, while closeness to kin held up across the same long gaps.

This is why "we’ll pick up where we left off" is reliably true of a sibling and unreliable of a friend. A sibling relationship has a structure underneath it that survives silence. A friendship is largely made of the contact itself, so when the contact goes, the thing goes with it — not immediately, and not dramatically, but steadily.

It also means the maintenance cost is not a personal failing. It is a property of the relationship type. Everyone pays it or loses the friendship, and most people are not told this in advance.

Not scheduling. A calendar reminder to "catch up with Sam" is an instruction to generate, from nothing, both the motivation and the content of a conversation — at whatever arbitrary moment the reminder fires. Most people dismiss it.

What has better evidence behind it is specificity. Deciding in advance when and where you will do something has a medium-to-large effect on whether you actually do it, measured across 94 tests and more than 8,000 people. "Message Sam" is a nag. "On the walk back from the shop, ask Sam whether the flat sale went through" is a plan, and it works because the walk does the remembering for you.

The other half is having something to say. A message that contains a specific thing you remembered is easier to send than a blank "hey, long time" — and it lands differently at the other end, because a remembered detail is evidence that you were paying attention, which is difficult to fake and impossible to generate.

Write down one specific thing within a few minutes of seeing someone. Not a summary of the conversation — one concrete fact with a date attached if it has one. "Second-round interview Thursday." "Mum’s surgery in March." "Dog had the operation, no running for six weeks."

Then make sure something brings it back to you before the next contact, rather than filing it somewhere you would have to remember to check. This is the part people skip, and it is the part that matters: an archive nobody revisits does nothing at all, which is the central finding of thirty years of lifelogging research.

None of this requires an app. A notes file with a name and a date works. What does not work is intending to remember, because that is the exact thing that fails.