We tracked 925 accounts, along with the legislatures, courts, and agencies around them, for hate aimed at Jewish, Muslim, and Christian people between January 1 and July 20, 2026. Widest is a count: 39 of the 59 confirmed items target Jewish people. Deepest is severity: every script that reached P1, our top tier, targets Muslims.
Most of it repeats a single narrative
26 of the 59 items run the identical story: a hidden, coordinated Jewish hand behind events with an unrelated real cause. Fuentes runs it for a million followers, Jones three times in three months, Owens for six million subscribers, and a candidate ran it inside California's state-mailed voter guide. None of them cite each other.
The most severe material came from institutions
The largest cluster is institutional: between February and June, the claim that Sharia law is an active governing threat moved through a House hearing, a Senate bill, a signed Florida law, a state attorney general's action, and a 94.8 percent party primary vote. Robinson's mass-deportation speech reached 36 million documented views.
Very little at the account level
Our dedicated 38-account anti-Christian lane produced nothing all half-year. The one account-level item came from ISIS's own spokesman. The anti-Christian hostility we could document appears in physical events, nine attacks and threats, mostly in Nigeria, and in institutional advocacy that stays inside protected speech.
Reach
These are the audience sizes of the accounts themselves, counted only where a direct, verifiable figure exists. When the only available number describes a news outlet that covered an item rather than the account that produced it, we leave it out of this list and label it as an outlet figure in the case files.
In their own words
The rates
Share of each category's membership-verified accounts that produced at least one confirmed item this half-year. The raw numbers sit beside each bar. Click a bar, or focus it and press Enter, to open that category's case files.
Severity
Each confirmed item is assigned to a narrative script. Each script is scored on reach, toxicity, coordination, mainstreaming, and harm potential. Click one for its scorecard.
The paper trail
Dated, sourced actions by legislatures, agencies, officials, and institutional actors. Click a row for the quote and source. Rows marked scored cleared our bar for a confirmed item; rows marked context are related events that did not.
Case files
Click a row for the quote, the coding, and the source.
Method
The roster holds 925 accounts across 23 ideological categories, each admitted through a documented selection procedure with two gates: a verifiable qualifying act or verified membership in the category's ideological current, plus independent standing. A confirmed account has a documented, dated act of hate on record, useful for tracking that account over time, and a poor basis for comparing categories, since it was pre-selected on the thing being measured. A lane account is verified as belonging to the category's ideological current, with no prior-hit requirement. Every comparative rate on this page uses lane accounts only.
Language that collectivizes, dehumanizes, or calls for exclusion of a religious group. We apply a stated-claim test: the implied claim has to be statable as "group are/do predicate." Criticizing a movement, a policy, a government, or a named person over their own conduct stays protected, in every lane, symmetrically. An official attacked over their exercise of official power reads as a proxy for their political camp, and language that does collectivizing work through that proxy is scored accordingly.
Five factors, each scored 1 to 5: reach (R), how many people the item plausibly got in front of; toxicity (T), how dehumanizing the language itself is; coordination (C), whether carriers are actually working together; mainstreaming risk (M), how close the item sits to institutions and large audiences; and harm potential (H), how directly it points toward real-world harm. They combine as (R·T·C·M1.3·H1.5)1/5.8: multiply all five, with mainstreaming risk raised to the 1.3 power and harm potential to the 1.5 power because those two carry the most real-world weight, then take the 5.8th root. The 5.8 is the sum of the weights (1 + 1 + 1 + 1.3 + 1.5), which brings the result back onto the 1-to-5 scale. Multiplying instead of adding means no single factor can carry a score alone. Priority tiers: P1 at 4.0 and above, P2 from 3.0, P3 from 2.0, P4 below.
Reach uses the first real number available, in order: a direct post metric, a show-average, domain traffic, follower count. Some figures measure the reporting outlet rather than the actor's own audience, and each case file labels which it has. Fuentes's follower figure traces to Wikipedia rather than a live platform pull. Three implausible domain figures were dropped rather than repaired. Where nothing citable existed, the item says so.
Every item rests on an independently fetchable, dated source read in full, never a headline or a screenshot taken on faith. Items that failed re-verification were removed entirely. Disclosed edges in this dataset: the Albanese statement rests on a single outlet so far; the Texas primary proposition is a non-binding party ballot measure, counted as part of a documented push rather than as law; the institutional scan concentrated on the United States and Nigeria, so its counts are not global.