Wire services and .gov own the facts. Think tanks own the framing.
-
01Wikipedia wikipedia.org
Baseline for officeholders, legislation, agencies, history.
T2Encyclopedic -
02Reuters reuters.com
Wire-service factual authority — global politics and policy.
T3Publisher -
03Associated Press apnews.com
Wire-service factual authority — US politics and policy.
T3Publisher -
04Federal & agency .gov sites various .gov
Primary source for legislation, rules, agency actions.
T1Government -
05New York Times nytimes.com
Editorial authority — frames the conversation.
T3Publisher -
06Washington Post washingtonpost.com
Editorial authority on federal politics and policy.
T3Publisher -
07BBC bbc.com
Global politics — geographic balance to US-heavy sources.
T3Publisher -
08Pew Research pewresearch.org
Polling, public opinion, demographic authority.
T1Academic -
09Ballotpedia ballotpedia.org
Election mechanics, candidates, ballot measures.
T2Encyclopedic -
10Brookings brookings.edu
Think-tank framing on policy and governance.
T1Academic
Opinion · attribution of intent · contested historical framing. Engines hedge. Think tanks fill the void. Framing is decided by which think tanks the engines retrieve.
Pairs with the shipped AI Policy Citation Share Study. Both halves of the same map — who is discussed (CSS) and who supplies the answer (this study).
- Which sources do AI engines cite most for politics?
- Reuters, AP, federal .gov sites, Wikipedia, NYT, WaPo, BBC, Pew, Ballotpedia, and Brookings. Wire services and .gov sources dominate factual prompts.
- Why does Ballotpedia rank highly in election answers?
- Structured, neutral, schema-tagged data on candidates, ballot measures, and election mechanics. No government source covers the same scope at the same depth.
- How do think tanks influence AI political answers?
- On opinion and "why did this happen" prompts where engines hedge factual claims, think-tank citations fill the void. Brookings, AEI, and Pew supply most framing-layer citations.
- Are wire services cited consistently across engines?
- Yes — AP and Reuters are the most consistent citations on factual prompts. Combined: roughly a quarter of observed public-affairs answers.
- Can political brands influence their AI citation share?
- Yes — but indirectly. The path runs through Wikipedia accuracy and earned coverage in trusted publications. PAC-owned domains rarely surface.
- Which political prompts have the most engine variance?
- Attribution of motive, contested historical framing, and polarized topics. Engines hedge differently and retrieve different think-tank balances.
Method
Citation share modeled across four AI engines — ChatGPT, Claude, Perplexity, Google AI Overviews — and a fixed prompt set of 60+ queries spanning informational, transactional, comparison, safety, "best of," and explanatory classes.
Sources tagged on the five-tier Retrieval Hierarchy: T1 Government & Academic · T2 Encyclopedic · T3 Publisher & Trade Press · T4 Community Platforms · T5 Brand-Owned. Estimates are directional and date-stamped.
