A conspiracy theory works by planting a question. You hear, somewhere, that the Rothschilds own the Federal Reserve, or that Jewish financiers run Wall Street, and the next time you are at a keyboard you type it in. Ask a search engine, and it mostly tells you the truth: across the top results for 14 such questions, 117 of 123 debunk the premise as an antisemitic myth or answer the literal question with plain fact, and exactly one affirms it. But a search engine only answers the question you bring it, and a conspiracy does not spread by waiting to be asked. Go to the place people are already standing, an online conspiracy community, and the same trope is affirmed about six and a half times as often as it is rejected, the mirror image of search, where it is rejected sixty times as often as it is affirmed.
This report measures the information environment around the conspiracy-of-power trope, the claim that a hidden financial cabal secretly controls money and governments, in the two places a person meets it: the search results they pull when they ask, and the community feed they scroll when they do not. It measures how each treats the trope, not the truth of the premise, which is not in question. The claims here are documented antisemitic canards.
Key Findings
- In the top of search, of 123 results across 14 gateway questions, 117 (95%) debunk the trope or answer the literal question factually, and just 1 affirms it; debunking outnumbers affirmation about 60 to 1 (Hanover Institute study, 2026, first-party).
- In an online conspiracy community, across 690 posts and top comments on the same trope, affirmation outnumbers debunking about 6.5 to 1 (300 affirm, 46 reject); the search ratio is inverted (study).
- The trope survives in coded form: 85% of the community affirmations name no Jews at all, invoking the Rothschilds, bankers, or globalists instead, the dogwhistle that slips past both the search-layer fact-checks and the platform’s own moderation (study). Coded antisemitism running far ahead of explicit is the documented pattern (Weinberg and colleagues, 2025, peer-reviewed).
- It is live, not archived: in this corpus the trope had reattached itself to a current 2025 news story within the sample window, surfacing in posts coded as affirming the same week the story broke (study).
- The one place the corner reaches the front is a leak through authority: a conspiracy document, submitted as a citizen exhibit to a state legislature and stored on its .gov archive, is the single result that affirms the trope in any top ten of search (study).
What does the web say when you ask?
Put the trope to a search engine as a plain question and the top results overwhelmingly correct it. Of 123 results, 60 debunk the premise as an antisemitic myth and 57 answer the literal question with fact, together 95% of everything returned; 5 frame the matter as contested and 1 affirms it.
The institutional web owns these results. The Anti-Defamation League, the World Jewish Congress, Wikipedia, the Southern Poverty Law Center, academic journals, and Jewish educational sites recur across question after question, and their pages exist precisely to name the conspiracy-of-power trope as a canard and explain it. Even Reddit, where 9 threads surfaced, holds the line: every one came from a sourced, moderated question-and-answer community, and none affirmed the trope. For the naive searcher, the front door is locked.
What does it say where you do not ask?
The same trope, in the feed of an online conspiracy community, is treated the opposite way. Across 690 posts and top comments on the conspiracy-of-power claim, 300 affirm it and 46 reject it, so affirmation runs roughly 6.5 to 1; in search the ratio was 1 to 60. Where the librarian corrects the premise, the community is built around accepting it.
The affirmations are not fringe whispers but the room’s consensus, carried on posts with thousands of upvotes, asserting dynastic financial control over wars, central banks, and the world’s money. The premise the search results spend their pages refuting is, here, simply the shared starting assumption.
Why does the debunking never reach it?
Because the version that lives in the corner is coded, and the fact-checks answer the explicit version. 85% of the community’s affirmations name no Jews at all; they say Rothschild, banker, globalist, or “the families,” the substituted words built to carry the meaning to an in-group while reading as ordinary economic complaint to everyone else. A page titled “the myth that Jews control the banks” does not intercept a post about the Rothschilds and the bankers, and neither does a moderation rule that removes slurs. The same coded layer that lets the trope evade detection, the documented pattern in which implicit antisemitism runs many times ahead of explicit (Weinberg and colleagues, 2025), is what lets it live untouched a click away from its own debunking.
It is also current. The trope in this corpus was not a dusty tract; it had attached itself to a current 2025 news story within the sample window, surfacing in posts coded as affirming in the same week the story broke. The conspiracy-of-power claim is not idle: its search demand was associated with registered hate-crime counts in four of five model iterations (Aziani and colleagues, 2025, preprint; the authors note most theories they tested showed no clear link and that no causal direction is established), and the Rothschild and Soros narratives carry rising search attention year over year.
Where does the corner reach the front?
Once, through authority. The single search result that affirms the trope in any top ten is not a conspiracy forum; it is a document on a US state legislature’s official archive, ranking third for “do the Rothschilds control the world economy.” It is a printout of a social-media post that quotes a conspiracy article stating, as fact, that the Rothschilds established the Federal Reserve in 1913 and still secretly control it. A member of the public submitted it as an exhibit to a committee hearing in 2023, where it was scanned into the record as every exhibit is. The legislature did not endorse it; the archive preserved it, as archives do. But a search engine reading a .gov address cannot tell a citizen’s exhibit from a state’s finding, and so the corner’s content inherits the front door’s authority and ranks just beneath the encyclopedia.
Methodology and limitations
This report presents an original Hanover Institute study and ships its full method.
Provenance. Two corpora on the same conspiracy-of-power trope. For search, a fixed set of 14 gateway questions, each the trope phrased as an ordinary query, was queried against Google from a United States location, capturing the top organic results, the follow-up question boxes, the presence of an AI-generated search summary, and any Reddit thread’s top comments. For the community, a fixed set of trope-term searches (Rothschild, central bankers, bankers’ wars, globalist control, “who controls the money”) was run against a large online conspiracy community, capturing the top posts of the past year and their top comments.
Sample. Search: 14 questions, 123 top organic results, 9 Reddit threads. Community: 125 posts and 690 coded units (posts and top comments). The unit of analysis is one result, post, or comment.
Enrichment. Each unit’s stance toward the trope premise was coded by a frontier-model pass as affirms, debunks, neutral, or, for search results, contested, from what a reader sees, and community affirmations were additionally coded explicit (naming Jews or Israel) or coded (Rothschild, banker, globalist). Because affirming a source of antisemitism is a serious label, every search result coded as affirming was hand-verified, which removed false positives where an academic source quoted the trope to analyze it, where a social post’s snippet showed a hateful comment rather than the post’s debunking message, and where a dual-meaning term was used in its political sense; the lone affirming search result was confirmed by reading the full document.
Limitations. Stance is coded from what a reader sees, not the full text of every page, the right unit for “what is each environment serving” but one that can miss a stance buried deeper. The community corpus is seeded by trope terms and skews toward engaged, upvoted content, so its ratios describe that conversation, not the platform as a whole; its raw affirmation rate is sensitive to how adjacent broad-conspiracy posts are counted, which is why the robust finding is the inversion of the affirm-to-debunk ratio between the two environments, not a single percentage. Both corpora are one search engine and one community, in one country and one window; results change. The single affirming search result is hand-verified, not model-certified. No private individual is named, and no community post is quoted; the recurring claims are characterized in aggregate.
Conclusion
So when a curious person, handed the question by a conspiracy theory, asks the web whether a hidden hand of bankers controls the money, what comes back? An honest answer, almost always: the top of search corrects the premise 95% of the time, and Reddit’s sourced communities correct it too. The trouble is that the person most exposed to the trope is not the one asking the question.
The two findings sit a click apart. Ask, and the top of search corrects the trope: 95% of 123 results debunk it or answer the question with fact, and 1 affirms it. Scroll the feed where the convinced already are, and the same trope is affirmed about six and a half times as often as it is rejected, with 85% of those affirmations naming no Jews at all, carried instead in the Rothschilds and bankers and globalists that a fact-check written against the explicit version is not built to catch. The single place the corner reaches the front is through authority, a citizen’s conspiracy exhibit ranked third for the weight of the .gov archive that stores it.
What the evidence settles is narrow and worth settling: a search engine is a poor instrument for catching a conspiracy, because it answers the few who ask and is silent for the many who already believe. What it leaves open is the harder question. If the correction waits behind a door only the doubtful knock on, and the trope lives, in coded form, in the feed the convinced are already scrolling, then what is a debunk-on-demand worth against an input whose search demand was associated, in one partial and correlational analysis, with registered hate-crime counts (Aziani and colleagues, 2025, preprint), and what would it take for the correction to reach the room where the trope is already at home?
Frequently Asked Questions
Are search engines reliable on antisemitic conspiracies?
For these 14 questions, largely yes: 95% of the top results debunk the premise or answer the literal question factually, and the institutional web dominates. But reliability for the person who asks is not the same as reach to the person who does not, and the same trope is affirmed roughly six and a half times as often as it is rejected in the community where it lives.
Why is the trope so much more common in the community than in search?
Two reasons. The community is built around accepting conspiracies, so the trope is its shared assumption rather than a question to answer; and the community speaks in code, naming Rothschilds and bankers rather than Jews, so neither the search-layer fact-checks nor the platform’s slur-based moderation intercepts it.
Does this mean Reddit spreads the conspiracy?
It depends on the room. The mainstream, moderated Reddit communities that surface in search debunked the trope, 0 of 9 affirmed it. The dedicated conspiracy community is the opposite environment. The same platform holds both the correction and the consensus.
What was the document on the government website?
A citizen exhibit, a printout of a conspiracy post that a member of the public submitted to a legislative committee hearing in 2023 and that was scanned into the public record. The legislature neither wrote nor endorsed it; a search engine ranked it third for the authority of the .gov domain that stores it.
Why does the conspiracy-of-power trope matter enough to measure?
It is among the oldest antisemitic tropes and a measured input, not an idle one: search demand for the Rothschild conspiracy has tracked registered hate-crime counts (Aziani and colleagues, 2025, preprint), and the most unmoderated fringe platforms carry antisemitic slurs and imagery at elevated rates (Zannettou and colleagues, 2020, peer-reviewed).
Sources
- Aziani, Lo Giudice, Yazdi, 2025. Conspiracy to Commit: Information Pollution, Artificial Intelligence, and Real-World Hate Crime. arXiv:2507.07884. Preprint.
- Hanover Institute, 2026. Study of the conspiracy-of-power trope across two environments: the top search results for 14 gateway questions, and the posts and comments of an online conspiracy community. First-party study.
- Google, 2026. Public organic search results, follow-up question boxes, AI-generated search summary presence, and search-volume data, United States locale. Public platform data.
- Weinberg, Levy, Edwards, Kopstein, Frey, and colleagues, 2025. Hidden in Plain Sight: Antisemitic Content in QAnon Subreddits. PLoS ONE. DOI 10.1371/journal.pone.0318988. Peer-reviewed.
- Zannettou, Finkelstein, Bradlyn, Blackburn, 2020. A Quantitative Approach to Understanding Online Antisemitism. ICWSM. DOI 10.1609/icwsm.v14i1.7343. Peer-reviewed.