Amy Weinkauf, who presented the KGI working group's "Better Feeds" findings to the commission, said recommender systems often start by winnowing millions of items into a ranked list based on signals such as past engagement, user profile and content characteristics. "The point of a recommender system is to help platforms sort through an enormous amount of content," she said.
Weinkauf warned that optimizing for short-term engagement — clicks, likes and watch time — can misrepresent users' long-term goals and amplify sensational or misleading content. "The potato chip problem" illustrates the mismatch, she said: behavioral signals don't always reflect what people truly want, and platforms that chase engagement can make matters worse.
The working group's recommendations fall into three categories. On design transparency, Weinkauf urged platforms to disclose the major categories of signals they use and their relative importance so external researchers and regulators can evaluate systems without exposing proprietary code. On user choice, she recommended platforms offer alternative recommender settings, including at least one explicitly optimized for long-term user value and defaults that protect young users. On evaluation, she said platforms should run long-term holdout experiments lasting at least 12 months, publish anonymized aggregate results and allow independent audits to validate whether changes improve user outcomes over time.
"Platforms should be evaluated based on long-term user value, not short-term engagement," Weinkauf told commissioners, adding that engagement metrics can still play a role if they reflect deliberate, not impulsive, behavior.
Weinkauf said KGI also produced a model legislative toolkit to help policymakers translate these recommendations into law. Commissioners raised enforcement questions; Weinkauf said KGI had not endorsed a single enforcement model and that the right approach would depend on context and policy goals.
The commission paused for questions and agreed to consider the report and model legislation as it develops recommendations for the legislature.