The reason a default sort order shapes the market

A marketplace sorts by relevance and the top three results take a large majority of the clicks. Sellers ranked fourth are not marginally less successful; they are in a different business. Nobody voted for that distribution, and the ranking that produced it was chosen by an engineer optimising a click metric.

What everyone sees

Ranking is read as a report on what buyers prefer, which makes high position a consequence of quality. The causal arrow runs both ways and the feedback is fast. Position generates the clicks that justify the position, so after a few cycles the ranking is largely reporting on its own earlier output.

What is actually happening

Ghose, Ipeirotis and Li examined ranking effects in an online marketplace and found that position substantially influenced demand independently of product characteristics, with re-ranking producing large shifts in what sold. Craswell, Zoeter, Taylor and Ramsey compared click position-bias models and found that click probability falls steeply with rank in a way that examination models capture better than relevance alone, meaning that clicks cannot be read as clean evidence of preference. Put together, a ranked list is a demand allocation mechanism whose outputs are then fed back in as inputs. In 2026, when the ranking is generated by a model trained on prior engagement, the loop is tighter and the original human choice about what to optimise is several retraining cycles out of sight.

Why it stays hidden

The circularity hides because the interface offers alternative sort options. The presence of sort by price or by rating implies that the default was one neutral choice among many, when in fact the vast majority of sessions never change it. The available alternatives are what make the default look like a preference rather than a decision.