This Answer Was Not Sponsored: And Why You Couldn't Tell If It Were

Advertising has boarded every medium — press, radio, TV, search — entering labeled, migrating into the substance, regulated one medium too late. LLM assistants are next, and this time influence vanishes into the model’s parameters, the weights, where no label or log reaches. Imagined as spycraft, a secret loyalty is an objective baked into those weights that serves whoever put it there, stays silent on ordinary questions, and denies itself when asked. The advertising business model produces the same structure with an everyday beneficiary, behind an “answer independence” pledge no outsider can verify, and needs no conspirator, only incentives and tools in print. We assemble them: a taxonomy of commercial secret loyalty, a stage model of the incentive gradient toward the weights, a 10,800-generation replication of the black-box audit (loud on a broad loyalty, silent on narrow ones), and a governance case for upstream data-provenance access, enforceable today through the industry’s pledges. Preprint. Originally developed for the Secret Loyalties Hackathon (Apart Research × Formation Research), July 2026; this version is a post-review revision. Code and data: https://github.com/4242labs/secret-loyalties