How Clipping File works

A search tool for the newspapers the Library of Congress has scanned, built so that a hit gives you the article instead of a whole broadsheet page to squint at.

Where the newspapers come from

Every page, scan and line of text here belongs to Chronicling America, the Library of Congress collection built with the National Endowment for the Humanities' National Digital Newspaper Program: over 20 million pages from US newspapers printed between 1770 and 1963. The pages are in the public domain. Clipping File doesn't copy or host them; it asks the Library for each piece when you search, and every clipping links back to the page at loc.gov.

Clipping File is not made by, or affiliated with, the Library of Congress.

What happens when you search

1. Your browser sends the search to the Library's public search service (the same one behind loc.gov) and gets back up to 20 matching pages at a time.
2. For each page it downloads the Library's OCR file, which lists every word the OCR found and where it sits on the scan.
3. It finds your words in that file. For an exact phrase it looks for the words in order; otherwise it picks the spot where most of your words sit close together.
4. It takes the column around that spot, reaches up to the headline when the headline is close, and asks the Library's image server for just that rectangle.
5. It shows about 80 words of OCR text either side of your words, and builds the citations from the page's catalogue record.
6. It compares the clippings with each other. When most of the short word runs in two clippings match, they're shown as one story printed in several papers, earliest first.

All of that runs in your browser. There's no Clipping File server between you and the Library, and nothing about your search is sent to us.

What it can't do

• The text is raw OCR. It's the Library's machine reading of century-old type, shown exactly as read: misspellings, broken words and stray characters included. Nothing is corrected, rewritten or summarized. Read the scan for the real wording, and quote from the scan.
• A new search is slow. The Library's search service often takes 10 to 40 seconds to answer a search it hasn't seen before. Once a search has run, the same search comes back in under a second (also for anyone you send the link to).
• The crop is a best guess. Some pages have one OCR block for a whole column; then the clipping is the stretch of column around your words, not exactly the article. The "Full page at LOC" link always opens the whole page.
• Spelling variants. The Library's search matches some forms of a word that the OCR text doesn't spell the same way. When the words can't be found on a page, that page is listed under "Every page" without a clipping.
• Rate limits. The Library allows a limited number of requests a minute. Clipping File stays well under them and, if the Library ever asks your browser to slow down, stops asking for an hour, as the Library's rules say.
• Coverage. Chronicling America is big but not complete: which towns and years exist depends on what each state's library scanned. Papers after 1963 aren't in it.

Free, and what will cost money

Searching is free, and so are the first 10 clippings of every search, with their citations and a sample PDF. The plan is a paid dossier PDF with every result of a search. It isn't on sale yet.

Who made it

Clipping File is made by aaon, a small software studio, and was built with AI assistance (the code, not the newspapers: there's no AI anywhere in what a search does). Questions or a clipping that came out wrong: aaonvsaaon@gmail.com.