How much did The New York Times write about each new mayor in his first 185 days, and how did it frame him? Zohran Mamdani, measured against Eric Adams, Bill de Blasio, Michael Bloomberg and Rudolph Giuliani at the same point in their terms.
From inauguration day through day 185. All mentions counts every article where the mayor's name appears. Featured keeps only the articles where he is front-and-center in the headline or lead — which also strips out false matches like Bloomberg the news company and other people named Adams. Featured share is measured by an AI read of a 240-article sample per mayor.
Weekly article counts across the first six months, every mayor started at the same clock. The honeymoon spike, and how fast it fades.
Bare surname counts (raw). Distinctive names (Mamdani, de Blasio, Giuliani) are effectively clean; Bloomberg and Adams curves include some company/namesake noise — see the relevance rates below.
Each relevant article's headline and lead, labeled by an AI classifier as positive, neutral or negative toward the mayor. Neutral is the default: straight, procedural news.
Tone of coverage is not the same as a mayor's standing. Labels are model-generated from headline + lead only, and coarse by design.
One newspaper, chosen because it is the only free source indexed uniformly from 1994 to today. It is a single, consistent yardstick; it does not measure all coverage.
Why only the Times. GDELT (TV + web) only reaches back to 2015; the TV News Archive to 2009; ProQuest/Factiva are licensed with no API. A five-mayor comparison back to Giuliani (1994) is only possible, for free and reproducibly, through the NYT Article Search API.
Counting. Articles matching each mayor's surname, bucketed by week over 185 days. We tried the cleaner person-facet and headline filters first — both are broken in the current API (they return zero) — and exact-name phrases, which undercount badly ("Michael Bloomberg" matched 1 article in two weeks; "Bloomberg" matched 175). So the surname search is the base.
Two metrics. "All mentions" is every article where the surname appears anywhere — the complete, directly-measured count, but for Bloomberg it includes the news company and for Adams other people named Adams. "Featured" keeps only articles where the mayor is front-and-center in the headline or lead. We measure the featured share by having Claude read a 240-article sample per mayor (headline + lead only) and judge whether the mayor is evident up top; featured total = all mentions × that share. The AI sees only the headline and lead, so featured excludes deep-body mentions; it is a strict lower bound on coverage, applied the same way to every mayor.
Sentiment. In the same pass, Claude labels each featured article's headline + lead as positive / negative / neutral toward the mayor, temperature 0. Coarse by design; every label is stored and auditable.
Confidence. Volume within scope: high. Cross-era volume: medium (32 years of indexing drift). Sentiment: medium-low.
Full write-up and scripts: METHODOLOGY.md.