How ProClose reads a PDF

Updated July 11, 2026

When you drop a carrier quote on the New proposal page, ProClose reads it right there, in your browser, before anything else happens.

#Where the reading happens

The parser runs entirely on your machine. A bundled PDF engine opens the file, walks every page, and pulls out the positioned text, no upload, no server round trip, no PDF ever leaves your computer to get read. This is the same promise that shows up on the Security page: your documents stay yours.

Under the hood there are two passes. The primary pass uses a full PDF text-layer reader that handles the vast majority of carrier-issued PDFs. If a page comes back with no extractable text at all, or the primary pass throws an error, a second, simpler reader steps in as a fallback, a hand-built scanner that looks for the raw text-drawing instructions inside the PDF's content stream. Between the two, most text-layer PDFs come through fine.

#Rules, not AI

Nothing about the reading step involves a language model. The parser is a set of deterministic rules, pattern matches and keyword lookups tuned against real carrier declarations pages, not a model guessing at what a field probably says. That matters for two reasons. First, the same PDF produces the same extracted fields every single time, there's no variance to account for. Second, nothing in your quote or policy PDF is sent to an AI provider, ever. The rules run locally and that's the end of the trip.

This is also why the parser skips boilerplate pages automatically. Carrier PDFs are full of legal notices, fraud warnings, and state-specific disclosures that carry no dollar figures and no coverage data. The parser recognizes these pages by their known legal language and skips over them when hunting for coverage fields, while still keeping them in the document for citation purposes. A real declarations or schedule page is never mistaken for boilerplate.

#What it's built for

ProClose's parser is built for text-layer PDFs, the kind that come straight out of a carrier's quoting system, where the text itself is embedded in the file and selectable. That covers the large majority of quotes an agency handles day to day: general liability and property packages, surplus lines quotes, monoline auto, and current policies pulled from a carrier portal.

Here's the honest capability line. ProClose reads real carrier PDFs, it is not a demo that only works on one canned sample. But it does not yet perfectly extract every field from every carrier's layout. Declarations pages vary a lot between carriers, and a layout the parser has never seen before might miss a field it would catch on a familiar one. That's exactly why every extracted number carries a citation back to the page it came from, see Page citations, so you can always verify what the parser found against the original document before anything goes in front of a client.

Preview note

Preview note: field-by-field extraction is proven and improving, but it is not guaranteed to catch every line on an arbitrary carrier's table layout today. Check the citations.

#What happens if a PDF can't be read

Not every PDF has a usable text layer, a scanned image with no embedded text, for instance, has nothing for either pass to read. When that happens, ProClose doesn't leave you staring at a blank screen. See When a PDF does not parse for exactly what the app does and what to try next.