How to search the Epstein files
The DOJ released over a million documents with no working search. Here is how to actually find something in them, including what the official tools cannot do.
The Department of Justice has released more than a million pages of Epstein-related material. Almost nobody has read any of it, and the reason is not indifference — it is that a release is not the same thing as access.
This is a practical guide to getting an answer out of that pile.
Why it is hard
Three problems, and they compound.
The documents are images. Much of what was released is scanned paper. A scan has no text layer, so pressing Ctrl-F does nothing. Until somebody runs optical character recognition over it, a million pages is a million pictures of pages.
There is no index. The files arrive as production batches with Bates numbers —
identifiers like EFTA00014187 stamped on each page for legal tracking. That numbering
tells you where a page sits in the production. It tells you nothing about what is on it.
The interfaces are not built for searching. Official reading rooms tend to serve documents one at a time behind a viewer that cannot be enumerated. You can read a document you already know about. Finding the one you did not know about is the actual problem.
The approaches that work
1. Search the text, not the filenames
The only thing that scales is full-text search over OCR'd text. Anything that requires you to guess a filename first is a browsing tool, not a search tool.
What to search for: names, obviously, but also places, dates, aircraft tail numbers, company names and phone numbers. Documents that never mention a person by name often place them somewhere, and a flight log or a message header is frequently more locating than a name in a body of text.
2. Search for the document type, not the topic
"Epstein trafficking" returns everything and therefore nothing. Searching the shape of the record works better:
- flight logs and passenger manifests
- deposition transcripts and exhibit lists
- visitor and call logs
- financial records, wire transfers, invoices
- correspondence between named parties
You are looking for a category of paper, then reading what is in it.
3. Follow the Bates numbers
Once you have one relevant page, its Bates number is a position in a sequence. The pages immediately before and after it are usually the rest of the same document — the other pages of the deposition, the following month of the log. A single hit is almost never the whole record, and the neighbours are the fastest lead you will get.
4. Read what the redactions imply
A redaction has a shape. A blacked-out span the width of two words in a list of names, on a page where the surrounding names are visible, is information about what was withheld even though it is not the withheld thing. Note them; do not infer identities from them.
What to be careful about
A name appearing in a document means the name appears in a document. Nothing more. These releases contain victims, witnesses, investigators, people mentioned in passing, people copied on an email chain they never replied to, and people accused of nothing whatsoever. Treating an appearance as an allegation is the single most common error made with this material, and it is how real people get accused of things no document says they did.
Check the document, not the summary. That includes ours. Anything worth repeating is worth opening the page for, and any tool that will not show you the page is asking you to trust it.
OCR is imperfect. Machine-read text from a poor scan drops characters and mangles names. If a search returns nothing for a name you expect, try a partial or a misspelling before concluding the record is silent.
Using this site
We index the DOJ releases and a number of other declassified collections, OCR'd and full-text searchable, and the assistant cites the file and page it took an answer from so you can open the original.
Two things worth knowing. It searches what has been released and ingested — an absence in the search is not evidence of absence in the world. And it will answer questions about documented abuse in these files without refusing or lecturing you, which is a deliberate decision: material this serious is not researchable through a tool that treats the question as the problem.