New York City Mayor Zohran Mamdani holds a press conference at the New York City Office of Emergency Management, as a major winter storm spreads across a large swath of the United States, in Brooklyn, New York City, US, Jan. 25, 2026. Photo: REUTERS/Bing Guan

When New York City Mayor Zohran Mamdani stood at Brooklyn’s Kings Theatre on June 18 and called AIPAC one of the “monsters” New Yorkers face — accusing the organization of moving “millions in dark money” toward “a single goal” of “turning us against one another” — most of the ensuing debate followed a predictable path.

Did he mean it as antisemitic? Was the criticism of AIPAC’s lobbying spending legitimate? Did Jewish leaders overreact? Within 72 hours, every major outlet had filed a version of the same story: a political speech, an accusation, a defense, a partisan split.

The intent debate is the wrong question — at least for understanding what happens next.

What matters analytically is what specific phrases do once they leave the stage: which phrases are taken up and reproduced by audiences, which are not, and what the propagation of conspiratorial language costs the public sphere when it occurs.

On this question — for the first time in this particular controversy — we now have data.

The Data

Mamdani used several distinct phrases at the rally. Two of them were taken up extensively by audiences and reproduced across reception environments. Two others received no observable audience uptake at all. This asymmetry is what makes the case analytically useful, and it is what the 2,396 coded comments allow us to document.

Over the five days following the rally, the Decoding Hate project at NYU’s Center for the Study of Antisemitism coded 2,396 audience comments across 24 posts and threads: 15 supportive Instagram posts on accounts amplifying the rally’s framing (1,497 comments), 5 critical Instagram posts diagnosing the language as antisemitic (499 comments), and 4 mainstream-press YouTube threads from Forbes Breaking News, CNN, NBC New York, and CBS New York (400 comments).

Each comment was coded against the Decoding Antisemitism Lexicon (Palgrave Macmillan, 2024), a published catalog of 40 historically documented antisemitic predicate-clusters and the phrases that signal them in contemporary discourse, developed over five years of empirical research.

The findings challenge the assumption that institutional critique stays institutional once it enters the digital reception environment.

In the supportive Instagram threads — accounts amplifying the rally’s framing — 75% of comments participated in the antisemitism-relevant reception ecology: 22% directly deployed antisemitic patterns (compression from “AIPAC” to “Jews” as a group, claims about Jewish control of US politics, dehumanizing imagery with documented Nazi-era genealogies), and a further 53% substantively endorsed those patterns without adding new ones.

Per-post combined rates ranged from 58% to 96%. The mainstream-press YouTube threads — the supposed moderating third space, where measured news coverage might temper the partisan extremes — produced an explicit antisemitic-trope activation rate of 42.5%, with a combined antisemitism-relevant rate of 80%. The four threads together drew approximately 309,000 views. Across 400 coded comments, no substantive anti-antisemitism counter-speech was observed.

Selective Propagation

Conspiratorial language does not travel uniformly. Some predicates carry an entire historical repertoire and propagate quickly through reception environments that already recognize them. Others — phrases that describe actions or motivations in more attributive form — produce no observed audience uptake at all. This study’s central finding is that the difference between the two is not random.

The word “monsters” was reproduced approximately 23 times across the most-engaged supportive posts and 30 times across the mainstream-press YouTube threads, extending from AIPAC toward Israel, Israelis, Zionists, and Jews as a group.

“Dark money” appeared in 98 coded comments across the YouTube threads alone — roughly a quarter of all coded comments — each time activating the broader Power and hidden-influence framing, extending the target to “Jewish billionaires,” “Jewish media,” and an entire register of claims about Jewish coordination of American political life.

Under one mainstream-press YouTube thread, a commenter wrote: “FINALLY, someone has the balls to call AIPAC the genocidal bankroll murderers they are” — 1,655 likes, the most-engaged single comment in the stratum.

Religious-evil escalation surfaced across the corpus — Satanyahoo as a portmanteau, demonic registers in multiple languages — and parasite imagery (parasite, tick, lice, leech) recurred across four different supportive accounts, each applied to Israel, Zionists, or a Jewish-coded target.

A single dehumanizing word against an institution multiplied across audience comment ecologies into an entire register of historical predicates.

The corpus’s most consequential single comment compressed all of this into two words: “Jewish playbook.” No mediating reference to AIPAC. No scaffolding. Just a coordinated-agency predicate attached to “Jewish” as a group-level attribute, with the commenter assuming the reader would track the implication. This is what audience appropriation of conspiratorial framing produces when it reaches its endpoint.

By contrast, two other phrases from the same rally — “single goal of turning us against one another” and “fear of peace” — did not trigger the kind of audience uptake that “monsters” and “dark money” produced. Neither phrase was reproduced, extended, or compressed into a group-level attribution anywhere in the 2,396 coded comments.

They describe motivations and actions in attributive form, without naming a moral antagonist in absolute terms or alleging corrupt power operating through hidden channels — and without those two operations, the historical repertoire that received “monsters” and “dark money” had nothing to recognize and nothing to extend.

Audiences are not passive recipients of speaker framing. They reconstruct political language using historical repertoires that already exist within the discourse environment, extending the predicates that environment equips them to extend.

The Audience-Side Repertoire

The findings should worry anyone who cares about the conditions of democratic deliberation, regardless of where they stand on AIPAC, Israel, or the Mamdani administration.

A century of historical scholarship has documented what happens when political language attributes hidden coordination, essential malign purpose, or dehumanized agency to a Jewish-associated target.

The predicates do not stay attached to the named institution. They are received through an audience-side repertoire — built across 19th-century European antisemitic discourse, Nazi propaganda, the Protocols of the Elders of Zion, and modern conspiracy traditions — that supplies the rest of the framework automatically.

The audience does not perform a logical inference from “AIPAC bad” to “Jews bad.” The repertoire was already there; the conspiratorial phrase simply activates it.

This is not a hypothetical concern. On the day of the rally, a Federal grand jury indicted a Florida man for an alleged December 2024 mass-shooting plot targeting AIPAC employees at the organization’s office. Two days earlier, Federal authorities had announced charges against five men in a separate alleged accelerationist plot whose target list included AIPAC-supported lawmakers.

The ADL’s Audit of Antisemitic Incidents 2025 recorded 1,160 antisemitic incidents in New York State alone, with New York City accounting for 860 of those — including 85 of the state’s 90 antisemitic assaults. Public speech about Jewish-associated organizations does not enter an empty environment. It enters one already saturated with the predicates it activates.

What our data shows is that this saturation operates most powerfully in the environments most receptive to the rally’s framing.

Supportive Instagram threads — accounts amplifying Mamdani’s framing — extended the rally’s predicates outward, from AIPAC toward Israel, toward Israelis, toward Jewish billionaires and Jewish media, and toward Jews as a group.

The mainstream-press YouTube threads, where audiences were supposed to encounter the controversy through measured journalistic framing, did the same: 42.5% of comments explicitly activated antisemitic predicates, and “dark money” alone appeared in 98 comments that opened the framing outward from AIPAC to broader claims about Jewish coordination of American political life.

The framing the speech introduced did not stay attached to the institution the speech named. It moved into the discourse environments that propagate it.

The Moderation That Didn’t Happen

The mainstream-press YouTube finding is the most institutionally consequential. The standard assumption among journalists, editors, and platform-policy professionals is that when news organizations cover a polarized controversy, their framing exerts a moderating influence on audience reception. Mainstream coverage, the theory goes, is the third option — the measured space between partisan amplification and partisan denunciation.

The data we collected — across four high-engagement threads from Forbes Breaking News, CNN, NBC New York, and CBS New York, with approximately 309,000 combined views — show that this assumption no longer describes contemporary digital reception environments.

The mainstream-press comment ecology around the Mamdani-AIPAC controversy produced antisemitism-relevant activation rates as high as, or higher than, the partisan-advocacy Instagram accounts. The moderation that mainstream framing was supposed to provide did not happen.

This is a platform-architectural finding as much as a content finding. Engagement-based algorithms reward emotionally provocative content. Recommendation systems create filter bubbles. Anonymity removes social accountability. Mutual reinforcement among co-located commenters produces escalation. In practice, this means that conspiratorial framings — once introduced by an authoritative speaker into the digital environment — propagate through reception ecologies whose architecture tends to amplify rather than temper them.

What Conspiratorial Language Costs

Political actors across the spectrum — left, right, or center — choosing words about Jewish-associated institutions, journalists framing controversies of this kind, and platforms shaping reception environments should weigh not only what language is intended to mean but also which historical repertoires that language is likely to activate in the audiences that receive it.

The substantive political critique of AIPAC’s spending, lobbying activity, and policy positions remains available — and legitimate — in language that does not overlap with antisemitic tropes. The same critique could have been made in other words. What our data show is that those specific words — not the critique itself — drove the audience-side activation, and that activation measurably degraded the discourse environment in which contemporary publics encounter one another.

That is the cost of conspiratorial language in democratic debate. It is borne not only by the named target and by Jewish communities, but by the public sphere itself.

Matthias J. Becker, PhD, is AddressHate Research Scholar at NYU’s Center for the Study of Antisemitism. He is the founder and lead of Decoding Antisemitism — now its successor project, Decoding Hate — Research Advisor to AddressHate, and Editor-in-Chief of Digital Hate Review.