← Back to Blog

Can You Train an AI Chatbot With a Scanned PDF or a Photo of Your Price List?

A huge number of small businesses in India don't have a clean, typed-up price list or menu sitting on a computer somewhere — they have a scanned copy of a printed sheet, or a photo taken on a phone. A common worry before signing up for any AI chatbot is whether that's even usable, or whether it only works with "proper" digital documents.

Why this trips up a lot of tools

Most document-based AI tools read a PDF by pulling out its text layer — the actual typed characters stored inside the file. A scanned PDF doesn't have one: it's really just a picture of a page saved as a PDF, so a tool that only looks for a text layer finds nothing at all, even though the page is full of readable information.

What happens when you upload one to Chatsettu

Your AI employee tries to read the normal text layer first, the same as with any PDF. If that comes back empty — the sign of a scanned or photographed page rather than a typed document — it automatically switches to reading the page visually instead, recognizing the text the way a person reading the photo would, in English or Hindi. You don't choose this yourself or configure anything; it happens automatically the moment a scanned document is detected.

A single photo of a menu or price list (uploaded as an image rather than a PDF) has worked this way for a while already — what's new is that the same recognition now also runs automatically on multi-page scanned PDFs, so a scanned 10-page catalog gets the same treatment as a single photo, page by page.

Where the honest limits are

This is genuinely useful, but it isn't unlimited. It's capped at 60 pages per document and a few minutes of processing time, so an extremely long scanned catalog may only get partly read rather than failing outright — whatever it manages to read in that time is still kept and used, nothing is thrown away. It also works best on a reasonably clear scan or photo; something genuinely blurry, at an angle, or handwritten will read less reliably than a properly scanned page, the same way it would for a person squinting at it.

Why this matters more than it sounds

The businesses who need this most are exactly the ones a lot of AI tools quietly assume don't exist — a neighbourhood shop with a photocopied rate card, a clinic with an old scanned brochure, a restaurant whose menu was only ever printed, never typed. Not needing to retype everything by hand before your AI employee can use it is often the difference between actually setting one up and giving up halfway through.