Hate posts first.
Before the attack there is almost always a thread. A Facebook group arguing about a zoning board. A comment section after a car crash. A word like "invasion" that keeps coming back. The research is clear that this chatter is not noise. It is the tremor before the quake, and it can be measured.
Illustration, not data. How a thread reads to an analyst: long stretches of grumbling, then a climb as the language turns from complaint to target.
Scroll or swipe sideways →
The proof is in the outages
Müller & Schwarz · Germany, 2015 to 2017Economists Karsten Müller and Carlo Schwarz took every recorded attack on refugees in Germany, about 3,300 of them across 4,466 towns, and lined them up against anti-refugee posts on Facebook.
When hostile posting spiked, attacks rose most in towns where more people used Facebook. In their working paper that came to roughly 35% more attacks in the high-exposure towns.
Then the clever part. Correlation is cheap, so they used the internet going down as a natural experiment. In weeks with a severe local internet outage, the link between posts and attacks disappeared. During Germany-wide Facebook outages, it weakened sharply. Pull the plug on the feed and the extra violence goes with it.
The effect showed up in arson, assault and property damage, not in peaceful demonstrations. Their companion study in the United States found the same pattern with Twitter and anti-Muslim hate crimes.
Roughly 40% fewer attacks during Germany-wide Facebook outages and roughly 70% fewer during severe local internet outages, relative to the mean. Estimates from the authors' working paper; the published version is in the Journal of the European Economic Association (2021).
Attackers leak
U.S. Secret Service, National Threat Assessment Center · 2016 to 2020The Secret Service studied 173 mass attacks in American public spaces. The finding that matters most to anyone watching a feed: attackers almost never come out of nowhere. Most of them worried somebody first. They posted, they talked, they made the people around them nervous.
Those 173 attacks killed 513 people and injured 1,234. Threat assessment exists because the warning signs were there. The job is making sure someone is reading them.
The ladder
How a thread climbsHate rarely jumps straight to a threat. It climbs. Every rung lowers the cost of the next one, and the crowd in the comments supplies the push.
So we don't just count ugly posts. Volume tells you a group is angry. Movement tells you it is getting dangerous. What we score is how fast a conversation is climbing, how many people are climbing with it, and whether it has stopped talking about "them" and started talking about a street, a building or a person.
The top two rungs are where a watch becomes a call to the people who can act.
Logistics
Times, routes, gear, "who's coming." Talk of being at the site.
Saturday, out front of the place. Bring a few guys.
Call to action
The crowd tells itself someone has to do something.
Somebody needs to handle this before the vote.
Targeting
Abstractions become addresses: a named rabbi, a school, a road, a council meeting.
The new building on Main, that's theirs.
Dehumanizing
Invasion, infestation, replacement. People recast as a threat to be repelled.
It's an invasion. They're taking over the town.
Grievance
Zoning, traffic, a crash, a school budget. Real complaints with a group attached.
Another accident. You know who was driving.
How it travels
From one post to a parking lotThe most dangerous threads are local. A national account ranting about Jews is everywhere and nowhere. A neighborhood group ranting about the Jews buying houses on one road knows exactly where to go.
The pattern repeats. A real local issue gets a group attached to it. The crowd rewards the angriest version. Someone screenshots it into a bigger room, outsiders pile in, and the thread comes home hotter than it left. Then it shows up in person.
That is why we watch where people live, not just where they shout.
The American numbers
ADL Audit of Antisemitic Incidents · FBI hate crime dataAntisemitic assault, harassment and vandalism recorded by the ADL. 2025's 6,274 is down 33% from the 2024 record and is still the third-highest year since the ADL began counting in 1979.
The ADL counted 6,274 antisemitic incidents in 2025, about 17 a day. New York led every state with 1,160. New Jersey was third with 687.
The FBI's own count of anti-Jewish hate crimes hit 2,321 in 2024, the highest since the bureau started collecting in 1991. Jews are about 2% of Americans and were the targets of roughly 70% of religion-based hate crimes.
A falling year is good news. It is not the all clear. The 2025 drop came mostly from campuses quieting down. Neighborhood hate, the kind that knows street names, did not get the memo.
How ANIA watches
Rules we work underPublic only
Open sources: public pages, public groups, public posts. Read-only. We never post, reply, bait or pose as anyone.
Speech is not a file
We follow the federal 28 CFR Part 23 standard voluntarily. Nobody gets a file for speech, politics or religion alone. Individuals are logged only when reasonably linked to criminal conduct.
A human signs off
Software flags the climb. A person reviews it. Nothing goes to police or a community without that sign-off, and referrals are for what police likely don't already have.
We watch hate against every community, not just our own. The same ladder that ends at a shul door ends at a mosque, a church or a refugee shelter. The ladder doesn't care who is at the top of it.
If you see it climbing
For community members- Screenshot it with the date, the group name and the link. Posts vanish fast.
- Don't argue in the thread. It feeds the pile-on and teaches them to move somewhere quieter.
- A threat, a named target or a plan goes to local police and the FBI at tips.fbi.gov. Call 911 if it's imminent.
- Tell your shul, school or community security lead so they can adjust before anything happens.
- Send it to us. Corrections and additions to anything on this page are welcome too.