Which charge is doing the work
Pick up to eight terms. Colour stays with a term while it is selected, so adding or removing one never repaints the rest. Dotted rules mark documented milestones.
Term frequency · Full range
Rising now
Momentum over the last three months against the preceding year. This is the “keeps up” view: which charge is gaining ground at this moment, rather than which one peaked historically. Click any row to chart it.
The vocabulary
Every term carries a direction of use: who deploys it against whom in the documented record — left→right, right→left, or contested by both. Several terms sit on the opposite side from where they are usually assumed to be. Click any card to add it to the chart.
Documented milestones
Dated, citable moments when a term crossed from a subculture into general political speech. Unlike the curves, these are factual claims.
Method, and what this cannot tell you
What is measured
The intended measure is the share of news coverage in which a term appears, monthly.
tools/fetch_gdelt.py pulls this from GDELT’s DOC 2.0 API, which indexes
online news from 2017 onward and needs no API key. Until you run it, the app ships modeled
curves and says so on every screen.
The reference study for the longer view is Rozado, Al-Gharbi & Halberstadt, “Prevalence of Prejudice-Denoting Words in News Media Discourse” (Social Science Computer Review, 2023) — 27 million articles from 47 US outlets, 1970–2019. It found a sharp and broad rise in prejudice-denoting terms across 2010–2019, accelerating after 2015. That finding is the empirical backbone of this tracker.
Counting a word is not reading it
This is the limitation that matters most, and the study’s authors say so themselves: frequency counts carry no context. An article about the overuse of “racist” counts the same as one deploying it. A rise can mean the accusation is being made more often, that the underlying conduct is being covered more, or that the word has become the subject of argument. PolitiFact’s review of a viral chart built on this data is a useful caution: the counts are real, the agenda read into them is not in the data.
Why terms broaden
The mechanism behind much of this is Nick Haslam’s “concept creep”: harm concepts expand outward to new phenomena and downward to milder ones. “Prejudice” once meant open antagonism; it grew to include implicit and unconscious attitudes toward a widening set of groups. Haslam notes the uncomfortable possibility that concepts broaden because the severe cases decline. “Gaslighting” in this dataset is the clean example.
Direction of use
Terms are tagged by who uses them against whom, because that assignment is usually assumed rather than checked — and the assumption is wrong often enough to matter. Vocabulary also changes hands: “fake news” was coined by journalists in late 2016 and captured within about ten weeks; “woke” and “cancel culture” both began inside Black vernacular English and now function as attack terms from the right. The sharpest case of a misplaced assumption is “groomer” and the associated pedophilia accusation: these are frequently taken to be part of the left’s vocabulary, but the documented record runs the other way. The term was seeded in a March 2020 4chan campaign and mainstreamed in March 2022 by Florida Gov. DeSantis’s press secretary, aimed at opponents of HB 1557. It is used predominantly by the right against LGBTQ people, teachers and Democrats — see GLAAD, the ISD explainer, and NPR/OPB. It is tagged right→left here for that reason.
A tool that tracked only one coalition would mostly measure its own assumptions, so the dataset covers all three directions. That is not a claim that the sides are symmetrical — the counts differ, the escalation curves differ, and you can read those differences off the chart. It is a claim that you cannot see any of that unless both are in the frame.
Editorial judgement
Cluster labels, direction tags and the milestone list are editorial calls, not outputs of
a model. They live in data/terms.json and data/events.json and are
meant to be argued with and edited.