Pink noise: conversations that drift then realign tend to be more enjoyable

Clinicians spend much of a visit in conversation, yet researchers still struggle to say what separates an exchange that clicks from one that doesn’t. Most accounts rest on similarity: partners who borrow each other’s words tend to fare better on outcomes ranging from romantic interest to cohesion in small groups. Those accounts score similarity as a single average for the whole exchange, which says little about how it changes from turn to turn.

A new analysis asks whether those fluctuations carry information of their own. It borrows the notion of pink noise, a pattern from signal processing in which small shifts are nested within larger ones across many timescales, and applies it to how closely each speaker's words track the other's. The analysis extends earlier research from a single similarity score to how similarity rises and falls during a conversation, using recorded conversations rather than laboratory experiments.

The rhythm of a conversation was linked to how much people enjoyed it, though the associations were modest. Across more than 1,600 conversations between strangers, higher pink noise scores accompanied higher enjoyment ratings and, to nearly the same degree, a stronger sense of connection. Personality ratings showed no link.

Two smaller datasets pointed in the same direction. Among fewer than 100 conversations between friends, pink noise again tracked enjoyment but not connection. The link to enjoyment did not hold once turn count and conversation length entered the model. Among 31 speed dates, those that ended in a mutual yes or maybe scored higher than those that did not. That held with conversation length in the model.

Pink noise also appeared to rise with practice. Across 25 podcasts, scores climbed over the first 50 episodes. Claude-written scripts for enjoyable conversations between strangers carried more of it than scripts for unpleasant ones, and human raters scored the enjoyable set much higher.

Pink noise did not beat every alternative. Once the more complex linear measures of similarity entered the model together, it no longer added explanatory power for enjoyment or connection. 

The main dataset, CANDOR, included 1,656 video chats between strangers, who rated their enjoyment and connection afterward. Smaller datasets covered friends, speed dates, and two-host podcasts, and Claude 3.7 Sonnet generated 100 coffee-shop scripts. Pink noise scores came from changes in similarity between adjacent turns, compared with shuffled versions of each conversation. Every analysis was observational, so the results show association rather than causation.

Showing that conversational rhythm shapes enjoyment would require experiments that deliberately alter how partners coordinate, which the authors say still have to be built. One early test would ask whether the practice effect depends on talking with the same partner over time or carries over to new ones. The authors also cite behavioral supports for people on the autism spectrum as a longer-term goal, though none have yet been designed or tested.

“Speakers remain aligned for a while, diverge as one introduces something new, then realign as the other picks it up and develops it further,” the authors write. 

Source: O’Nell KC, et al. (2026 Nov 6) Sci Adv. Pink noise in dyads’ semantic similarity trajectories predicts conversation enjoyment