Original experiments on how AI assistants choose which brands to mention, cite and recommend. Each one is pre-registered: design and hypotheses are fixed before any answer is generated, and deviations are reported.