How to convert a PDF to text with AI (including scans)

The right way to pull text out of a PDF depends entirely on one thing: whether the file already contains real text or is just a picture of a page. A text PDF gives up its words in seconds with a copy or an export. A scanned PDF needs optical character recognition (OCR), and modern AI tools now do that job well. Here is how to tell the two apart and pick the fastest method for each.

First, figure out which kind of PDF you have#

Open the file and try to select a sentence with your cursor, or press Ctrl+F (Windows) or Cmd+F (Mac) and search for a word you can clearly see on the page. If the text highlights or the search jumps straight to it, the PDF has a real text layer — it was created digitally, exported from Word, a browser’s “Print to PDF,” or an email client. If nothing selects and the search finds nothing, you are looking at a scanned PDF: a photograph of each page with no text underneath.

That single test decides the whole job. Text PDFs are trivial. Scanned PDFs, including phone photos of documents and anything that came off a copier or fax, need OCR to rebuild the words from the image.

The easy case: copy or export a text PDF#

If your file passed the selection test, do not overthink it. Select the passage you want, press Ctrl+C, and paste it wherever you need it, or use Ctrl+A to grab the whole page at once. For a long document, export the entire thing instead: most PDF viewers offer Export or Save As to a Word document or a plain-text .txt file, which preserves reading order better than copying page by page. If you want a fully editable file rather than raw text, our guide to convert a PDF to Word walks through the cleanest options.

AI chatbots also read text PDFs instantly, because the text layer is already there for them to parse. You only need the OCR methods below when there is no text layer to begin with.

Scanned PDFs: let AI do the OCR#

This is where AI genuinely helps. Instead of buying dedicated software, you can hand a scan to a general chatbot and ask for the words back.

• ChatGPT. Attach the PDF with the paperclip icon and ask something direct like Extract all the text from this PDF, preserving line breaks. Its vision models read the page as an image, so they cope with messy scans and even handwriting. Per OpenAI’s file uploads FAQ, the 512 MB per-file limit applies on every plan; the real difference is how often you can upload — free accounts are capped at a few files a day, paid plans allow many more.
• Claude. Upload the PDF at the chat window and ask for the text the same way. Anthropic’s PDF support documentation explains that Claude processes each page as both an image and a text stream and merges the results, which is why it handles scans and charts. Both tools slip on very small, blurry, or crooked text, so always proofread names and numbers.
• Google Docs OCR (free). Upload the scan to Google Drive, right-click it, and choose Open with > Google Docs. Google runs OCR automatically and drops the extracted text below a copy of the image; then use File > Download to save it as .txt or Word. Google’s convert files to text help notes the file should be “2 MB or smaller,” “right-side up,” and sharp, and warns that “lists, tables, columns, footnotes, and endnotes are not likely to be detected.”

Adobe Acrobat and dedicated OCR tools#

For long documents where you want to keep the original layout, Adobe Acrobat Pro is still the workhorse. Open the scan and, in versions after October 2023, choose All tools > Scan & OCR, then under Recognize text select In this file and press Recognize text, as Adobe describes in its OCR walkthrough. The page text becomes selectable and searchable in place; from there you can copy it or use Convert / Export to save a Word or plain-text version. This runs locally on the desktop app and requires a paid Acrobat Pro license. Acrobat’s separate AI Assistant is an extra-cost add-on aimed at summaries and questions, not raw text extraction, so it is the wrong tool for a straight OCR job.

If you need a free, offline option, dedicated OCR engines like Tesseract, or “copy text from picture” features built into note apps, do the job without sending anything to the cloud.

Method	Best for	Cost	Runs where
Copy / Export	PDFs that already have a text layer	Free	On your device
ChatGPT or Claude	Messy scans, handwriting, mixed layouts	Free tier plus paid plans	Cloud upload
Google Docs OCR	Quick, free OCR of clean single scans	Free	Google’s cloud
Acrobat Recognize Text	Long documents, keeping layout	Acrobat Pro (paid)	On your device
Tesseract / offline OCR	Sensitive files that cannot leave the PC	Free	On your device

A word on privacy before you upload#

Every cloud method above sends your file off your device. That is fine for a recipe or a public flyer, and a real problem for medical records, contracts, tax forms, or anything holding Social Security or bank numbers. Consumer ChatGPT and Claude can also use your conversations to train future models unless you opt out — in ChatGPT, turn off model training under Settings > Data Controls or use a temporary chat; Claude has an equivalent setting. Business, enterprise, and API tiers generally do not train on your content. The same caution applies when you summarize a PDF with AI. For anything truly sensitive, use local OCR — Acrobat’s desktop Recognize Text or an offline engine keeps the document on your machine — or redact the private parts first.

Getting clean text and next steps#

Start with the two-second selection test every time, because it saves you from running OCR on a file that never needed it. Text PDF: copy or export. Scanned PDF: match the tool to the stakes — Google Docs or a chatbot for clean, non-sensitive scans, and Acrobat or an offline engine for long or private documents. Whatever you use, read the output against the original, since OCR quietly drops digits, swaps similar letters, and scrambles tables. And if a large scan bumps into an upload limit, split the PDF into smaller parts or compress the file first, then run the text extraction on each piece.