Learn › AI Basics for Beginners › When AI Can See: Vision and Multimodal AI

When AI Can See: Vision and Multimodal AI

Beginner 🕐 7 min Lesson 7 of 13
What you'll learn
  • Understand the difference between AI that generates images and AI that reads them
  • Know how to send an image or document to a chatbot
  • Know the strengths of each major chatbot for vision tasks
  • Identify practical everyday uses for vision AI

Two Very Different Things

The previous lesson covered AI that generates images — you describe something and it creates a picture. This lesson covers something completely different: AI that reads images you send it.

Every major chatbot — ChatGPT, Claude, and Gemini — can now accept images, screenshots, PDFs, and documents as inputs. You hand them something visual and ask a question about it. This is called vision AI or multimodal AI (because it works across multiple types of input, not just text).

What You Can Send

Most chatbots now accept:

  • Photos — snap a picture of anything and ask about it
  • Screenshots — useful for troubleshooting software problems
  • PDFs and documents — upload a report, contract, or research paper and ask it to summarise or pull out specific information
  • Charts and graphs — ask the AI to explain what a chart is showing or extract the underlying numbers
  • Handwritten notes — works well for clear handwriting; gets less reliable with messy scrawl

How to Use It

Look for the attachment or paperclip icon in the chat interface. In ChatGPT, Claude, and Gemini, it appears near the prompt input box. Click it, upload your file or image, and then type your question as you normally would. There is nothing special about the prompt — just ask what you would ask a human who was looking at the same thing.

Some examples of things you can type after uploading:

  • "What does this chart tell me about sales trends?"
  • "Summarise this PDF in five bullet points."
  • "I got this error message on my screen. What is causing it and how do I fix it?"
  • "Pull out all the line items and totals from this invoice."
  • "What are the main arguments in this document?"

Which Tool Is Best for This?

All three major chatbots support vision, but they have different strengths:

  • Claude — particularly strong at reading long documents and PDFs with detailed layouts. Handles up to 100 pages natively and extracts text from visual elements like tables and diagrams.
  • Gemini — the broadest input support. Can handle text, images, audio, and video within a single conversation, and has a very large context window for long documents.
  • ChatGPT — convenient because image upload is built into the same interface most people already use. Handles PDFs by reading them as images, which works well for most use cases.

For everyday tasks — a PDF you want summarised, a photo you want explained, a screenshot you want help with — any of the three will do the job. Claude and Gemini edge ahead for complex or lengthy documents.

Real-World Use Cases

Here are situations where vision AI pays off immediately:

  • Receipts and invoices — photograph an expense receipt and ask it to extract the date, vendor, and amount for your records
  • Technical troubleshooting — take a screenshot of an error message and ask what is wrong
  • Research papers — upload a dense academic PDF and ask for a plain-English summary
  • Menu reading — travelling somewhere with an unfamiliar language? Photograph the menu and ask for translations
  • Whiteboard notes — photograph a whiteboard from a meeting and ask for a structured summary

What It Struggles With

Vision AI is impressive but not perfect. It tends to struggle with:

  • Very small or low-resolution text
  • Messy handwriting
  • Complex multi-column layouts where reading order matters
  • Images where important detail is obscured or blurry

When accuracy really matters — a legal contract, a medical report — always verify what the AI extracted against the original. Use it to save time on a first pass, not as a substitute for careful reading.

The Takeaway

You no longer need to retype or describe things to get AI help with them. If you can see it, you can send it. That simple shift — from describing to showing — unlocks a whole category of tasks that chatbots could not handle even a year or two ago.

Key takeaways
  • Vision AI reads images you send it — completely different from image generation
  • Every major chatbot (ChatGPT, Claude, Gemini) now accepts image and document uploads
  • Use the attachment icon in the chat interface and then ask your question normally
  • Claude and Gemini edge ahead for long or complex documents; ChatGPT works well for everyday use
  • Always verify important extracted information against the original source
Learn more every week Get 5 hand-picked prompts + new AI tools every Monday. Free.