Claude AI can now analyze PDFs - here's how to try it (and why you'll want to)

Adrienne Bresnahan/Getty Images

Anthropic has added a new feature to its Claude 3.5 Sonnet artificial intelligence (AI) model: Visual PDFs, or the ability to view and analyze PDF files (a feature OpenAI's ChatGPT already supports). The model can not only interpret text in a PDF but also images, charts, and graphs.

However, there's a catch -- the capability is available only through a paid professional subscription or the API. If you're not already a subscriber, here's why you may want to upgrade.

Also: Anthropic warns of AI catastrophe if governments don't regulate in 18 months

According to Anthropic's user guide, the new Claude 3.5 Sonnet works with any standard PDF, with a few restrictions. The PDF can't be larger than 32MG or more than 100 pages, nor can it be encrypted or password protected. To help Claude better analyze and understand a PDF, you'll want to make sure the text is clear and legible, contains standard fonts, and has its pages in the proper orientation. 

Screenshot by Lance Whitney/ZDNET

You can ask Claude questions about any text, pictures, charts, and tables in the PDFs you upload. Some examples given by Anthropic include:

Analyzing financial reports and the charts or tables they containExtracting key details from legal documentsTranslating documents into another languageConverting documents into more structured and organized formats

Anthropic's user guide explains how to use the new PDF analysis in the Messages API. That's fine for programmers and others who create apps using Claude AI, but what about the rest of us? That's where you'd need a subscription. 

Also: How the 2024 US presidential election will determine tech's future

For $20 a month, a Pro plan gives you early access to new features, among other perks. And in this case, the PDF analysis is a new feature, currently in beta mode.

How to access Claude 3.5 Sonnet's PDF feature

If you have a Pro plan, click your account name or email address at the bottom of the left sidebar and select Feature Preview from the menu. In the Feature Preview window, select Visual PDFs and turn on its switch.

Return to the main screen and start a new chat. Click the paperclip icon and choose a PDF from your computer. After the file's thumbnail appears under the prompt, type and submit your question or request.

Also: I tested 7 AI content detectors - they're getting dramatically better at identifying plagiarism

You can start by asking Claude to do something simple, like summarize the file. You can then move on to submit specific questions about the text in the file.

From there, try asking it about a specific image, table, or chart. To do this, refer to the page number that contains the image or chart. When doing this, use the logical number (the number displayed by your PDF viewer) and not the physical page number (the number displayed on the page). In Adobe Reader, for example, hover over the physical page number on the right, and the logical page number appears.

If a PDF file is too large for Claude to swallow, you can split it into smaller files and upload each one separately. On Anthropic's GitHub page, you'll also find a few sample PDFs that you can download and submit to Claude for analysis and answers.