American Factbook: a reference of U.S. public data, one page per agency, county, court and organization

Federal spending, crime, immigration, nonprofits, drug-company payments to doctors and farm subsidies, with every figure traced to the government dataset it came from. Free, no ads, built and maintained by an AI agent.

Most public data in the United States is published. Very little of it is readable. USAspending holds every federal award; the FBI publishes crime by agency; the IRS releases the filings of every tax-exempt organization; CMS reports what drug and device makers pay every doctor; USDA lists every farm program payment by payee. Each one is a portal, an API or a set of large files, organised the way the publisher organises it rather than the way a person asks a question.

American Factbook takes those datasets and turns them into pages: one for each agency, state, county, city, court, judge, nonprofit, doctor and farm-subsidy recipient the data describes. As of September 2026 that is about 2.4 million pages across six topics, and the site is built and kept current entirely by Claude and Claude Code, with a human editor reviewing the work.

What is on it

Federal spending. Federal agencies reported $5.3 trillion in obligations for fiscal year 2025 across 73 top-tier agencies. The hub has a page for every agency, every state, the largest recipients and contractors, counties, congressional districts and budget functions, plus the debt, outlays and the federal workforce. One finding from the data: North Dakota records more federal obligations per resident than any other state, $123,560, and 92.9% of it comes from one agency, because place of performance is where a program is administered rather than where the money lands.

Crime. The FBI's reported violent crime rate for 2025 is 329.3 per 100,000 residents, 58.7% below its 1991 peak. Every state, 1,826 cities and 929 counties have pages, alongside offense pages, prison and jail populations, and rankings that carry the FBI's own caution about ranking jurisdictions and show each agency's reporting coverage beside its rate.

Immigration. CBP recorded 34,428 nationwide encounters in August 2026, 90.7% below the December 2023 peak. The hub covers every sector and field office, and the immigration courts: each court with its completed and pending cases and asylum grant rate, and each of the 1,540 judges.

Nonprofits. The IRS recognises 1,964,958 tax-exempt organisations; 164,036 have a Form 990 on record. Organisations with $1 million or more in income have their own pages (Kaiser Foundation Health Plan, for instance, with $82.5 billion of revenue for 2024), along with rankings by revenue, assets and compensation, and every state.

Health. Drug and device makers reported paying $17.3 billion in general payments to 1.69 million doctors and other practitioners between 2019 and 2025, from CMS's Open Payments data, plus what each prescriber billed Medicare Part D. Every doctor with a payment on record has a page. The hub answers questions like how much drug companies pay doctors with the figure in the first sentence.

Farm subsidies. The Farm Service Agency paid 1,030,325 recipients $24.4 billion in 2025 under 71 programs, and the site holds every year USDA has published, 2004 to 2025, 77 million payments in all. The largest 10% of recipients took 64.2% of the money. There is a page per recipient, per state, per county office and per program, rankings, and questions such as how much your county received.

A state entry pulls the records for one state together: crime rate and rank, federal obligations per resident, nonprofits, doctors, courts. The full index lists every page type with its count.

How it is built, and why that matters

Every number on the site is the output of a query checked into the repository and run against a published dataset. The page shows the publisher, the dataset, the licence and the date the data was retrieved, and links to the source. Nothing is estimated, and no figure is written by hand: the text on a page is generated from the query's result, and a check fails the build if a printed figure does not match its value.

That discipline extends to what the site refuses to do. Person-level pages are the official record and nothing more; a doctor's page says what companies reported paying, not what the doctor earned or why. A farm recipient is a name and a mailing address exactly as USDA printed it, and two rows are the same recipient only when they match exactly, because USDA publishes no identifier. Where a dataset cannot support a claim, the page says so in a caveat above the figures rather than a footnote below them. Rankings carry the coverage column that makes them honest: a state whose police agencies cover 65% of its population is not compared as if they covered all of it.

When something is wrong, it is fixed and logged on a public corrections page, with the cause. The methodology explains each hub's figures, and the data register lists every dataset with its licence and how to cite it.

Who it is for

Anyone who needs a public number and would rather not read a federal API to get it: reporters checking a figure, local officials comparing their county with its neighbours, researchers, students, and residents who want to know what their state's crime rate is, what federal money is spent there, or which farms and doctors in their county appear in the record. The site is free, has no ads, and is not affiliated with any government agency.

It is also, quietly, an experiment in what an AI agent can maintain: six hubs, sixteen government sources, a nightly audit of the live site, and a corrections log, run by a coding agent under a written set of rules about what a number is allowed to claim. The rules, the queries and the decisions are all in the open, which is the only way a reference built this way can be trusted.

The site is at americanfactbook.org. New datasets and corrections are announced in its changes feed.