Behind the Database
How the Human Impact Project database is built
About This Database
The Human Impact Project is a living database documenting reported incidents of harm related to U.S. immigration enforcement. This database undercounts the true scale of immigration enforcement's impact, as it can only capture stories that have been publicly reported: people who have spoken to journalists or lawyers, or who have friends or family advocating on their behalf. The vast majority of detentions, deportations, family separations, and other enforcement actions go unreported entirely. Fear of retaliation silences some, while others are deported before they have the chance to make their story known. Local news deserts mean entire regions go uncovered.
Despite these limitations, the Human Impact Project's goal is to document as many individual stories as possible and centralize them in one searchable place, so that journalists, policymakers, researchers, and the public can see the human impact of immigration enforcement at the individual level as well as patterns at the regional and national level.
The Human Impact Project is a project of the Human Security Initiative, a 501(c)(3) nonprofit organization committed to research and advocacy promoting safety and dignity for all. Learn more at humsi.org.
Where the Data Comes From
The database draws from a wide range of publicly available sources:
National and local news outlets: Articles from newspapers, television stations, and digital news organizations across the country reporting on specific enforcement incidents.
Investigative and nonprofit reporting: In-depth reporting from organizations like ProPublica, The Marshall Project, ACLU, and other investigative and advocacy organizations.
Legal and court documents: Court filings, judicial decisions, and legal actions related to immigration enforcement.
Social media: First-person accounts, community reports, and video documentation shared on platforms like Instagram, TikTok, and Facebook, which often capture incidents before traditional media arrives.
GoFundMe campaigns: Fundraising pages created by families and communities affected by enforcement actions, which often contain detailed personal narratives.
An automated daily search uses the Exa news search API to find recent articles about ICE enforcement incidents across hundreds of news sources. Anyone can also manually submit a URL through the site's submission page.
The Role of AI
AI plays a specific, bounded role in the data pipeline — it assists with extraction and structuring, but does not make editorial decisions.
When a news article or source URL enters the system, the following happens:
• The article's text is extracted and cleaned using Mozilla Readability (the same technology behind Firefox's Reader View) to isolate the article body from ads, navigation, and other page clutter.
• The cleaned text is sent to an AI model (Anthropic's Claude) which extracts structured data: a headline, date, location, summary, incident type tags, and country of origin. The AI follows detailed extraction guidelines that prioritize leading with the human story, using neutral language, and presenting only the facts.
• If an article covers multiple distinct individuals or incidents (a "roundup" article), the AI identifies and separates each person's story into its own database entry so that each human impact story can be found independently.
• When multiple sources cover the same incident, the AI synthesizes them into a single cohesive narrative with a timeline, verifying that sources are actually about the same event before combining them.
• Locations mentioned in articles are geocoded (converted to map coordinates) so incidents can be displayed on the interactive map.
• Every source URL is preserved — through the Internet Archive's Wayback Machine where possible, or otherwise as a copy saved to the project's own storage — so that the underlying reporting remains accessible even if the original article is later edited, paywalled, or taken offline.
A team member reviews each entry before it is published to the database, verifying accuracy and correcting any errors in the AI's extraction. The AI does NOT decide what stories to include or exclude, does not editorialize or add commentary, and does not assess the credibility of sources. It is a structured data extraction tool.
Known Limitations
Undercount: As stated above, this database can only include stories that have been reported publicly. The true scale of enforcement activity is far larger than what appears here.
English-language bias: The automated search and most sources are in English. Reporting in other languages is underrepresented.
Geographic gaps: Areas with fewer local news outlets or less activist presence are underrepresented. Rural enforcement actions are particularly likely to go unreported.
Timeliness: There is an inherent delay between when an enforcement action occurs and when it is reported, processed, reviewed, and published in this database.
AI extraction errors: While the AI extraction is generally accurate, it can occasionally misidentify locations, dates, or incident types. Human review catches most of these errors, but some may persist.
Other Resources
This database focuses on individual human impact stories. For broader statistical data and aggregate numbers on immigration enforcement, the following resources are valuable:
Detention Reports: Tracks ICE detention facility data, including population counts, facility conditions, and detention statistics across the country.
TRAC Immigration: Syracuse University's Transactional Records Access Clearinghouse provides detailed statistical analysis of immigration enforcement, court backlogs, and detention trends.
Immigration Policy Tracking Project: A searchable compilation of Trump administration immigration policies, indexed by subject, agency, and type of action, with source documents and litigation status.
These resources complement the Human Impact Project by providing the aggregate data and policy analysis that contextualize the individual stories documented here.
Questions & Contact
For questions about the methodology, data, or any aspect of the Human Impact Project, please contact:
We welcome feedback, corrections, and tips on stories that should be included in the database.