When the Microphones Started Saying ‘We’: The Quiet Data Signal Behind Zohran Mamdani’s NYC Upset
A 300-hour transcript analysis reveals a year-long surge in collective pronouns and progressive policy talk—pointing to Mamdani’s victory well before polling did.
On Tuesday, first-term Assembly-member Zohran Mamdani stunned New York’s political class by winning the Democratic mayoral primary. His platform — a pro-tenant, pro-public-bank agenda delivered in unmistakably movement language — looked radical next to the cautious centrism of Cuomo, and other candidates in the race. I wanted to take a deeper look at this through the lens of City Council, Planning Commission and School Board meetings to see if there were early signs that this might happen. While it shocked the political world, it was not so surprising seeing that this could be a possibility based on what was happening in the public discourse.
The evidence in two charts
Collective-to-Individual Rhetoric (Fig. 1 above).
The ratio of “we / our / us” to “I / me / my” hovers just above parity through 2023, dips briefly, then explodes to 1.66 in 2025 Q2. When civic discourse frames issues collectively, campaigns built on solidarity — “we build power together” — find unusually fertile ground.
Progressive Keyword Density (above).
Mentions of a 20-term progressive lexicon (“housing justice”, “green new deal”, “public bank”, etc.) per 1 000 words climb modestly for five quarters, then quadruple in the year before the vote. The right-hand bars (Frustration Index) confirm the spike occurs alongside rising anger, not complacency — an electorate seeking change, not merely optimism.
Taken together, the charts show a discourse that first collectivises, then amplifies progressive policy talk — the precise rhetorical mix Mamdani owns.
How the analysis was done
Corpus & pre-processing. 151 meeting transcripts, 3.1 million tokens. Each file’s date → quarter; quarters with < 3 meetings were excluded to avoid thin data.
Normalization. All counts expressed per 1 000 tokens so a short sparse quarter cannot imitate a low-signal one.
Signals.
Collective ratio = count(we|our|us) ÷ count(I|me|my).
Progressive density = frequency of 20 hand-curated policy phrases validated against a BERTopic pass; lexicon chosen for clarity and auditability.
Frustration index = 12-word lexicon, plotted on a secondary axis.
Robustness checks. Removing the three longest meetings moves any quarterly value by < 0.02; bootstrapped 95 % CIs on the ratio never cross 1.0 after 2023 Q2; Spearman ρ between token volume and either headline metric < 0.12 — confirming trends are content-, not length-driven.
Why it is directionally correct, despite caveats
Speaker attribution blind-spot. Public-comment speakers and councillors are blended; future work could weight elected officials. Mitigation: the collective-pronoun surge is visible in both scripted (“agenda items”) and unscripted (“open comment”) sections, signalling it is city-wide, not limited to activists at the microphone.
Transcription noise. AssemblyAI’s word-error-rate on US council audio is ~6 %. Because both numerator and denominator of each metric are single words or short phrases, mis-recognitions dilute signals symmetrically rather than skewing them.
Lexicon scope. Slang terms (“NIMBYism”, “YIMBY”) were added manually after a first pass; still, jargon drift may under-count in later quarters. Benchmarking against BERTopic topics shows < 5 % of progressive-policy clusters lacked a matching lexicon term.
It’s not perfect, but it’s not bad. I’ll work on doing speaker identification for future posts.
Additional take-aways
Messaging fit beats media volume. Mamdani’s name appears in only 2.4 % of meetings, yet the ideas he champions flood the agenda — an early warning that headline-level sentiment tracking would miss.
Collective language as a leading indicator. The pronoun ratio flipped five quarters before the primary; similar flips preceded insurgent wins in Boston ’21 and Chicago ’23. Track it, and you get a one-year lead on polling.
Frustration as accelerant, not cause. Anger rose with, not before, the progressive surge; voters had already chosen a direction, frustration merely added urgency.
Conclusion
The transcripts show that by mid-2024 New York’s civic conversation had already turned plural (“we”) and policy-left (“housing justice”, “public bank”). Anyone mapping discourse, not polls, could have pencilled in a high-ceiling progressive and seen Mamdani as the only candidate who matched the moment. The lesson for 2026 and beyond is blunt: if council microphones start saying “we” and “justice” in the same breath, a movement candidate isn’t an upset — it’s the base case.