For most of recorded history in most cultures, the dominant route into a romantic partnership ran through the friend-of-a-friend pipeline. Your sister's classmate. Your cousin's coworker. Your roommate's brother. Your best friend's neighbor at the wedding. The pipeline was so dominant that it required no name, no infrastructure, no app. It was simply how meeting happened — a passive function of being embedded in a dense, overlapping social network whose members were continuously introducing peripheral nodes to each other for reasons that included but were not limited to romance.
Michael Rosenfeld's longitudinal data captures the pipeline's specific collapse. "Met through friends" was the single largest channel of American couple-formation from at least the 1940s through the 2000s, peaking at roughly 33% of all new couples in the mid-1990s. By 2017 it had fallen to about 20%. By 2022, with the post-pandemic apps consolidating, it sat below 15%. The friend pipeline did not slowly attenuate — it collapsed in the same fifteen-year window that the apps rose, and the collapse was not coincidental. The apps did not merely add a new channel; they substituted directly for the friend introduction, which was an almost identical service performed by a different intermediary.
Why does the substitution matter? Because the friend intermediary and the app algorithm perform the same function — pre-screening candidates and reducing search cost — through entirely different mechanisms. The friend pre-screens on character knowledge, life trajectory, value compatibility, and stake in the outcome. Your friend will not introduce you to someone they think will treat you badly, because they will face you afterward. The friend has skin in the game. The algorithm pre-screens on demographic correlates, declared preferences, and engagement patterns. The algorithm has no skin in the game; its reward function is your continued use of the app, which is poorly correlated with the quality of any specific match.
The collective consequence of pipeline collapse is that the population's average match-quality screen has been delegated from accountable peers to unaccountable software. Marie Bergström's "privatization of intimacy" argument applies precisely here: the friend-of-a-friend channel was inherently social, because the friend witnessed both parties and bore reputational consequences. The app channel is inherently private, because no observer is present and no reputational consequence attaches. A culture moving from one to the other is not making a neutral technological substitution; it is removing community accountability from partner selection at population scale.
The pipeline also performed functions invisible to the participants. Friends introducing friends produced cross-network bridges, weak-tie strengthening in Granovetter's sense, expanded romantic options for people whose own networks were thin, and a continuous low-grade circulation of social information that maintained community cohesion. The decline of the pipeline is therefore not just a romantic decline but a social-capital decline, of a piece with Putnam's broader documentation. Bowling leagues did not produce couples directly, but they produced the friends who introduced the couples. When the leagues died, the pipeline thinned, and the algorithm filled the gap.
There is a generational asymmetry here. Older Americans who came of age before 2010 retain functional friend pipelines and use them, often in parallel with apps. Younger Americans, particularly those who left college after 2015, often do not have a friend pipeline to use, because their friend networks are themselves thinner than prior cohorts'. The decline of friendship itself — documented in repeated Pew and Gallup surveys showing American adults reporting fewer close friends than at any previously measured point — feeds the romantic story. You cannot use the friend pipeline if you do not have friends close enough to know what you are looking for in a partner. The romantic crisis and the friendship crisis are not separate.
Reviving the pipeline is harder than reviving any single venue. Third places can be rebuilt by zoning and capital. The friend pipeline requires dense, overlapping, multi-decade friendships, which require third places plus stable geographic residence plus shared institutions plus time. Americans move more, change jobs more, live alone more, and spend less time with friends than they did in 1990. The pipeline is downstream of all of this. It cannot be turned back on by deciding, individually, to ask friends for introductions. It can only be partially restored by the slow rebuilding of the broader social infrastructure friendships depend on.
The Law 2 question is whether the substitution of algorithm for friend produces outcomes good enough to justify the loss. The available evidence is mixed. App-formed marriages are not obviously less durable than friend-formed marriages on the short evidence available; some studies suggest the opposite. But durability is not the only metric. The friend pipeline produced partnerships embedded in shared social networks from the start, which had downstream effects on family formation, child-rearing support, and community membership. The app produces partnerships that begin outside any shared network and have to construct one, which is harder and is often not done. The marriages that result may last but are more isolated. The cost shows up not in the couple but in the community around them.