Skip to content

EveryShot Methodology

EveryShot is a publicly available data tool released in March, 2025 that tracks gun violence incidents in the United States. It uses AI to parse news stories about gun violence. Below is a description of the methodology behind the database that powers EveryShot and detailed information about variable definitions.

Data Source

EveryShot data is sourced from news stories, so the data is as complete as the available news reports on shooting incidents. Not every incident is reported, and some communities and geographic areas have few or no media outlets reporting on local issues. Published articles also often lack detailed information about shooting circumstances and victim and shooter demographics.

As you use the tool, remember that a missing variable in EveryShot means that no media outlet reported that detail, not that it isn’t relevant or applicable. For example, the “handgun” filter only displays shootings where a handgun was mentioned in news reporting. It is not a record of every incident involving a handgun.

Obtaining and Verifying Articles

EveryShot uses Event Registry’s news aggregation service newsapi.ai as its principal source. Using select keywords, an initial pool of incidents is created. ChatGPT-4o mini then identifies articles of shootings that involve a real gun or firearm that was fired, or brandished and that occurred in one of the 50 US states, not including US territories. It excludes articles of shootings involving a toy, pellet, or BB gun and those that are about a fictional event, movie, or video game.

Creating Records and Assigning Geographic Information

ChatGPT-4o mini then ascertains whether each news story contains information about a single shooting or multiple shootings and assigns a single record for each separate incident.

The next step is assigning values for a set of variables (see Appendix A). For geographic information, if the city is known, ChatGPT-4o matches it against a database of unique latitude and longitude coordinates. If not, locations are assigned using Google’s geocode API and the latitude and longitude are added to each record.

Deduplication and Finalizing Records

Multiple news stories are often written about a single shooting event. In order to obtain only one record per shooting incident, we consider information about location, names, addresses, and date. Once stories about each incident are grouped together, ChatGPT-4o mini uses the top 10 stories or fewer per incident and their variable outputs in order to build as comprehensive a record as possible for each incident and to flag any discrepancies. These final variable outputs and corresponding articles for each incident are then reprocessed through ChatGPT-4o, a more powerful AI model, in order to verify accuracy. An additional layer of deduplication is conducted at this point to check victim/suspect names across a broader date range. This step helps identify and add articles pertinent to trials or charges arising later than the incident date.

Human Review

When the AI identifies discrepancies or uncertainty at any time during processing, the event is flagged for human review by Everytown staff.

Publication Process

Data is collected, processed, vetted, and pushed live to EveryShot twice weekly. Incidents are updated if additional articles are published within 30 days of the incident date.

Terms of Use

Certain features of EveryShot utilize and incorporate artificial intelligence, machine learning, and related technologies, including without limitation via features and functionalities licensed from or otherwise provided by third parties. Users of EveryShot acknowledge that AI is a new and developing technology, and that the outputs generated by AI features of EveryShot may be incomplete, inaccurate, offensive, or otherwise unpredictable. Everytown expressly and specifically disclaims any liability, responsibility, and/or guarantee regarding the outputs or processes of AI features on EveryShot. Use of EveryShot is entirely at the user’s own risk.

Appendix A: Variables

The following are the variables in EveryShot. Race categories used are American Indian / Alaska Native, Asian American / Pacific Islander, Black, Latinx, and white. We do not assume race if it is not explicitly stated.

Shooting Outcome

  • All Incidents
  • Fatal Shootings
  • Non-Fatal Shootings
  • Other

Intent

  • Gun Homicides
  • Gun Assaults
  • Gun Suicides
  • Attempted Gun Self-Harm
  • Shootings by Police
  • Shootings of Police
  • Unintentional Shootings
  • Gun Discharges – Other
  • Brandishing

Incident Type

  • Active Public Mass Shooting
  • Escalating Altercation
  • Extremism
  • Hate Crime
  • Home Invasion
  • Intimate Partner
  • LGBTQ+
  • Mass Shooting
  • Murder-Suicide
  • Political Violence
  • Road Rage
  • Trans-Involved
  • Workplace Shooting
  • Not Reported

Location

  • School
    • Elementary School
    • Middle School
    • High School
    • College or University
    • Other
  • Bar/Club
  • Home
  • Park
  • Business
  • Hospital
  • Indoors
  • Street/Sidewalk
  • Outdoor Event
  • Daytime event (7am–10pm)
  • Nighttime event (10pm–7am)
  • Not Reported

Victim

  • Gender
    • Male
    • Female
    • Not Reported
  • Race/Ethnicity
    • American Indian/Alaska Native
    • Asian American/Pacific Islander
    • Black
    • Latinx
    • White
    • Other
    • Not Reported
  • Age
    • 0–9
    • 10–17
    • 18–24
    • 25–64
    • 65+
    • Adult (no age specified)
    • Child/Adolescent/Teen (no age specified)
    • Not Reported
  • Relationship to Shooter
    • Self
    • Parent
    • Sibling
    • Friend
    • Intimate Partner
    • Other Family–Child or teen
    • Other Family–Adult
    • Bystander
    • Not Reported

Firearm

  • Firearm Type
    • Assault Weapon
    • Machine Gun
    • Handgun
    • Rifle
    • Shotgun
  • Firearm Ownership
    • Legal
    • Illegal
    • Unauthorized
    • Not Reported
  • Ghost gun (or unserialized)
    • Yes
    • Yes – 3D-Printed
    • No
    • Not Reported
  • Firearm Manufacturer
    • Remington
    • Springfield Armory
    • Glock
    • Smith & Wesson
    • Beretta
    • FN
    • Ruger
    • Jennings
    • Palmetto State Armory
    • SCCY, Colt
    • Taurus Arms
    • Sig Sauer
    • Taurus
    • Bersa
    • Colt
    • SpringField Armory
    • Ruger, Glock
    • Sig
    • KelTec
    • Mossberg
    • Stoeger
    • Walther
    • North American Arms
    • Radical Firearms
    • Cande Mefe Century Arms
    • Cobra
    • Diamondback
    • Springfield
    • Rockland USA
    • SCCY
    • Rock Island Firearm, Bersa
    • Taurus, Glock
    • Bear River
    • Phoenix Arms, Walther
    • Hi-Point
    • RIA
    • Llama
    • Canik
    • Kahr
    • Winchester
    • SIG Sauer
    • Harrington and Richardson Arms Co.
    • Sauer
    • Windham
    • Heritage
    • Walther, Ruger
    • Not Reported
  • Firearm Owner
    • Shooter
    • Shooter Family Member
    • Other
    • Not Reported
  • Stolen gun
    • Yes
    • No
    • Not Reported
  • High-Capacity Magazine
    • Yes
    • No
  • Machine Gun Conversion (e.g. “Glock switch”) used
    • Yes
    • No
    • Not Reported

Suspect

  • Gender
    • Male
    • Female
    • Not Reported
  • Race/Ethnicity
    • American Indian/Alaska Native
    • Asian American/Pacific Islander
    • Black
    • Latinx
    • White
    • Not Reported
  • Age
    • 0–9
    • 10–17
    • 18–24
    • 25–64
    • 65+
    • Adult (no age specified)
    • Child/Adolescent/Teen (no age specified)
    • Not Reported
  • Number of Shooters, 1–9