New AI Searches Long PDFs Up to 10x Faster

If you've ever tried to find one fact in a 200-page report, you know how painful document search can be. Most AI search tools handle this by scanning each page separately, which is slow and expensive. Worse, if the answer you need is spread across several pages, page-by-page searching often misses it entirely.

A team of researchers built a solution called DocPC. Instead of examining every page, the system picks the most important pages of a document and combines them into one simple grid image. The AI then searches that combined image as a whole. This approach reduces the number of images the system has to store and process by 10.1 times, and cuts preparation time by about 7.7 times. Imagine a document that used to take 10 minutes to get ready for searching now taking about 1.3 minutes.

It's also more accurate. Because the AI can see multiple key pages at once, it can connect information that spans different pages. In tests, DocPC outperformed traditional page-by-page systems on a new benchmark designed to measure multi-page relevance. The researchers made their code and data public, so other teams can build on the method.

The catch? It's still early-stage research, not a polished app you can use today. The system's performance depends on correctly choosing which pages are "representative" — pick the wrong pages and search results could suffer. But the speed and storage savings suggest that future document search tools could be significantly faster and cheaper for businesses, researchers, and anyone with a massive pile of PDFs.