This report examines a paradox at the center of modern governance: some of the most consequential policy decisions in democratic societies are shaped by institutions that no one elects and few people can name. Think tanks — RAND, Brookings, Heritage, AEI, CAP, CSIS, and their counterparts abroad — occupy a space between academia and government, translating technical complexity into guidance that legislators, regulators, and executives can act on quickly.
The report's central argument is that this influence operates through ideas rather than formal authority: agenda-setting, framing, legislative drafting, media amplification, and the movement of personnel between research institutions and government offices. But it insists on holding that argument to real evidence rather than asserting it in the abstract, and the evidence complicates the clean version of the story at every turn.
The opening chapter defines what a think tank actually does and where different institutions sit on a spectrum from "explain the world" to "change the world" — a distinction that matters more than any single label. It then works through the specific mechanisms of influence (agenda-setting, framing, drafting, media relationships, staffing pipelines, coalition networks) before turning to how political parties actively use, filter, and sometimes discard think tank research rather than adopting it wholesale.
A chapter on funding is the book's evidentiary core. Using three documented cases — Brookings' fourteen-year funding relationship with Qatar and the subsequent FBI investigation into its former president; the American Enterprise Institute's ExxonMobil-linked climate positions; and the Center for American Progress's undisclosed corporate "Business Alliance" — the book distinguishes between funding that shapes what gets studied, funding that shapes how it gets framed, and funding that shapes specific conclusions or suppressed criticism. It argues that institutional reassurances about donor independence are nearly impossible for outsiders to verify, while disclosure is checkable — which is why transparency, not trust, is the realistic remedy.
Four real-world case studies then ground the theory in checkable outcomes: RAND's documented role in drafting the 2023 Biden AI executive order (and that order's reversal fifteen months later); Brookings' diffuse, uncodified influence on AI regulatory debates; CSIS's contribution to the CHIPS and Science Act; and Heritage's "Mandate for Leadership," where more than 270 Project 2025 proposals were matched to Trump administration actions and named authors moved directly into the offices they had written policy for. Each case shows influence arriving bundled with friction — funding ties, personnel overlaps, and political liability that intensify in proportion to how effective the influence becomes.A forecast chapter resists the temptation to predict simple, ever-expanding influence. It identifies AI governance, biosecurity, and climate policy as likely growth areas, but weighs that against two countervailing forces: AI tools that may erode think tanks' core synthesis function, and funding scrutiny that scales with visible effectiveness. A synthesis chapter then draws the threads together, arguing that benefits and risks are not separate ledgers but the same mechanism viewed from two angles — the expertise that makes an institution valuable is typically funded by the same relationships that later invite scrutiny.The report closes by asking what could actually be done — granular donor disclosure, structural firewalls between research and fundraising, mandatory methodology publication, cooling-off periods on the revolving door — and why each proposal runs into real constitutional and practical resistance rather than simple inertia. Its final position is deliberately unresolved: think tank influence will likely remain real, consequential, and only partially accountable, producing not a crisis and not a fix, but a persistent, low-grade tension between expertise democracies need and oversight democracies haven't yet built.