When Big Data Enables Behavioral Manipulation

Acemoglu and co-authors build a formal model of something practitioners intuit but rarely see rigorously: platforms do not merely recommend products, they learn which product attributes deceive users most efficiently and exploit that knowledge dynamically.

The key concept — 'glossiness,' attributes that make a product appear better than it is — gives a precise vocabulary to what is otherwise discussed as dark patterns or attention manipulation.

The welfare analysis is the paper's sharpest contribution: AI helps consumers when deception is short-lived (the market corrects quickly), but when manipulation is durable, AI becomes an instrument against the user it appears to serve.

For product directors, this is an uncomfortable mirror — the same personalization infrastructure that improves recommendations also optimizes for the platform's interest when those interests diverge from user welfare.

Read alongside information economics classics in the library and against Sinan Aral's work on social influence: together they show how the platform layer captures value by shaping the epistemic environment users navigate.