# How to paste a screenshot into ChatGPT or Claude > Copy the screenshot to the clipboard (⌃⌘⇧4 on Mac, Win+Shift+S on Windows), click the chat box, and paste with ⌘V or Ctrl+V. Then ask your question. Take the screenshot straight to the clipboard, then paste it into the message box. On a Mac, press **⌃⌘⇧4** (the usual **⌘⇧4** plus Control) and drag over the area you want; on Windows, press **Win+Shift+S** and select an area, and Snipping Tool copies it for you. Then click into the ChatGPT or Claude message box, press **⌘V** (Mac) or **Ctrl+V** (Windows), type your question and send. It works the same in the web versions and the desktop apps. If you already have the screenshot as a file, drag it into the chat window instead. ## The built-in way ### Mac 1. Press **⌃⌘⇧4** and drag over the part of the screen you want. Adding Control sends the shot to the clipboard instead of saving a file. (**⌃⌘⇧3** does the same for the whole screen.) 2. Switch to ChatGPT or Claude and click in the message box. 3. Press **⌘V**. The image attaches to your message. 4. Type the question and send. If you took a normal **⌘⇧4** shot, you can also drag the floating thumbnail from the corner of the screen straight into the chat window before it disappears. ### Windows 1. Press **Win+Shift+S** (on Windows 11, **PrtScn** opens the same tool by default) and drag over an area. 2. The snip is on the clipboard. On Windows 11 it's also saved to **Pictures\Screenshots**. 3. Click in the message box, press **Ctrl+V**, add your question and send. ### Where it falls short The built-in tools are fast, and for "what is this error?" they're enough. The trouble starts when you want the model to look at something specific. A clipboard screenshot has no arrows, no numbers, nothing hidden: you can describe "the second button from the left in the toolbar" in words, or open the image, mark it up, save it and drag the file in. And anything on screen goes along with it: names, emails, the tab next to the one you meant. ## Why a tight, marked-up crop gets better answers - **Less to read means less to misread.** A whole-screen capture asks the model to figure out which of forty things you mean. A crop of the dialog that's broken doesn't. - **Big images get shrunk.** AI chat apps scale large images down before the model sees them; Anthropic's documentation, for one, describes images above a size limit being downscaled. A full-screen capture of a large display loses small text in that squeeze. A tight crop keeps it legible. - **Arrows and numbers are cheaper than sentences.** "Why is **1** misaligned with **2**?" beats a paragraph of coordinates, and the model reads the markers the same way a person would. - **What you hide can't leak.** Redact the customer's name before the screenshot goes into a chat history you don't control. ## With hypershot hypershot treats "get this into the chat" as the main event, not an afterthought. **Fastest: straight to the clipboard.** Set **Settings → Captures → After capture** to **Only Copy**. Every capture, say a region with `⌘⇧2` on macOS or `Ctrl+Shift+3` on Windows, goes straight to the clipboard; paste it into the chat. A small pill at the bottom of the screen, there for a few seconds, lets you open that capture in the editor if you decide it needs a mark-up after all. **Or drag it in.** With the default **Preview** setting, each capture lands as a card in the preview folder in the corner of your screen. Drag the card into the ChatGPT or Claude window, or into a coding agent's prompt if it takes images. It's a native drag, so it drops like a file from Finder or Explorer. The card's **Copy** button works too. *A capture dragged from the preview folder into an AI agent's prompt, then the question typed after it.* **When it needs pointing at:** open the capture in the editor and - draw an **Arrow** (**A**) or drop numbered **Counters** (**C**) on the parts you're asking about ([arrow](/docs/hypershot/tools/arrow/), [counter](/docs/hypershot/tools/counter/)); - **Redact** (**X**) anything personal, with **Secure** for a solid block ([redact](/docs/hypershot/tools/redact/), and [why blur isn't always enough](/blog/redact-before-send/)); - crop tighter with the **Artboard** tool (**B**). A region capture keeps the whole display it was taken on, so you can also drag the artboard back out if you cropped too tight ([artboard](/docs/hypershot/editing/artboard/)); - click **Copy** in the export pane, and paste into the chat. One thing to know: in the editor, **⌘C** / **Ctrl+C** copies the *selected annotations*, not the image. The **Copy** button in the export pane is what puts the whole picture, annotations included, on the clipboard as a PNG ([copy to clipboard](/docs/hypershot/exporting/copy-to-clipboard/)). **When it's only text,** send text. For an error log or a stack trace, `⌘⇧O` on macOS or `Ctrl+Shift+O` on Windows reads the text out of the selected area and puts it on the clipboard, which is smaller, searchable and quotable in the model's reply ([the OCR trick](/blog/the-ocr-trick/)). ## Questions people ask ### Can I paste more than one screenshot into a single message? Yes, in both ChatGPT and Claude: paste or drop them one after another before you send. With hypershot, numbering them with Counters helps you refer to them ("in the second screenshot, **3**…"). ### Why does paste do nothing? The clipboard probably holds text, not an image. On a Mac, make sure you used **⌃⌘⇧4** (with Control), not **⌘⇧4**, which saves a file instead. In hypershot, use the export pane's **Copy** button rather than **⌘C** / **Ctrl+C**. --- From https://hypernuclear.com/blog/paste-screenshot-into-chatgpt-claude/