Real emergency alerts from US colleges and universities, preserved word for word
The exact words universities choose when lives are at stake
An archive of real campus emergency alerts, each preserved in its exact, original wording and linked to a citable source. Free to search and export.
3,249 alert messages from 578 colleges and universities, documenting 1,252 incidents from 1969 to 2026. 2,680 messages (82%) are preserved word for word from a cited source; the remainder are documented in each case's timeline (type, channel, timing) without displayed text, because their exact wording is not preserved in the public record.
Archive last updated July 2026 · Full statistics →
Take the data with you
Download it all, then analyze it your way
Every case, every sequenced alert, every citation, free to export in one click. Pull the whole archive here, or filter to just the slice you want on Search the Archive, then drop it into Microsoft Copilot, Google NotebookLM, ChatGPT, Claude, a notebook, or a spreadsheet and ask your own questions.
- The verbatim text of every alert (2,680 confirmed word-for-word, 3,249 messages in all), one row per message.
- A source URL for every alert, plus channel, timestamp, confidence, casualties, and Clery category, so anything is checkable.
- Choose your depth: a quick case summary, the full verbatim-alert grain, or everything, in CSV, Excel, JSON, PDF, or HTML. Versioned downloads, the codebook, and citations live on the Dataset page.
1,252 cases · archive and exports: no signup · public-domain sources
Heads-up: the full archive is a large file (the complete JSON is tens of MB). The “Everything” and JSON/Excel options are biggest; CSV and the case-summary level are much lighter, and filtering to a subset on Search keeps downloads small.
The “Everything” level also folds in Claude’s element analysis (consensus and arbitrator synthesis) on each coded first alert, and the JSON export can optionally include all 25 individual AI reads. It is AI-generated and not human-reviewed.
One click = the whole archive, one row per alert. These are large files (tens of MB; JSON and Excel are biggest, CSV is lighter), so you’ll see the exact size and can confirm before it saves. Want a smaller subset or a different level of detail? Use Export or Search the Archive.
New here? How to analyze it
Export a file, upload it to an AI tool, then ask away.
Suggestions for AI tools that can read your export: Microsoft Copilot, Google NotebookLM, ChatGPT, Claude. Upload the file you downloaded, paste one of these starters, and go from there:
“I've uploaded a dataset of real, verbatim US campus emergency alerts. Walk me through the common phrasing patterns institutions use in the very first message of an active-threat alert, with real examples.”
“Using only the verbatim alert texts in this file, show me how all-clear messages are typically worded, and quote a few real examples.”
“From these real alerts, help me draft a clear, plain-language shelter-in-place notification, grounded in how actual campuses worded theirs.”
These are just starting points. The data is yours to question however you like.
Recently added cases
Recently added
Why this exists
Humans set the standard. AI did the work.
This started as a way to learn to use AI well. Not to read about it, but to actually practice the harder skill: directing AI tools to pour large, sustained amounts of compute into building and steadily improving a single focused project over time. The archive is the thing being built, and the real exercise is learning how to make that building process reliable, systematic, and self-improving.