Microsoft's new Click To Do overlay, now rolling out to Copilot+ PCs, turns any on-screen content into actionable items—summarizing text, extracting emails, or searching visually—using a mix of on-device AI and the cloud. But the feature's real-world usefulness depends on which CPU is inside your laptop, and it arrives with a trail of privacy and hardware compatibility fine print you should read before relying on it.
How It Works: From Screenshot to Action in Three Steps
Click To Do is fundamentally a screen-capture shortcut with AI smarts baked in. Invoke it by holding the Windows key and clicking your left mouse button (or pressing Windows+Q, or using the Snipping Tool), and Windows takes a full-screen screenshot. Optical character recognition (OCR) kicks in to identify text—emails, URLs, chunks of prose—while image analysis models detect objects, backgrounds, and visual elements.
From there, a context menu appears with actions tailored to what you selected. Pick an email address, and you'll see "Send email" to open your default mail client. Highlight a URL, and "Open website" launches your default browser (yes, it respects your browser choice). Select more than 10 words of text, and the AI-heavy options appear: Summarize, Rewrite (in casual, formal, or refined tones), Create a bulleted list, or even Ask Copilot for deeper analysis.
Under the hood, those intelligent text actions run on your PC’s NPU using Microsoft’s Phi Silica small language model. That means summaries, rewrites, and list creation happen locally—no data sent to the cloud, and responses stream quickly with minimal power draw. But when you choose cloud-based actions like Visual Search with Bing, Search the web, or Ask Copilot, only then does Click To Do route your content out to Microsoft services.
What You'll Actually Use It For (and What You Won't)
After testing early builds, the most practical use cases for Click To Do fall into three buckets.
1. Rapid text triage. You're staring at a long email or a dense article and want the gist without opening another app. Click To Do lets you select a paragraph, hit Summarize, and get a bullet-point synopsis in seconds. It's not as deep as a cloud LLM’s summary, but for quick extractions—to-dos from meeting notes, key points from a support page—it saves real time.
2. Visual search without context switching. Spot a lamp in a photo online? Select it, click Visual Search with Bing, and get shopping results without ever leaving the page. This is the kind of micro-efficiency that, if you remember it exists, shaves minutes off daily browsing.
3. Image fixes on the fly. Blur a background, erase an object, or remove a background from a screenshot directly from the overlay. These actions call into Photos or Paint, but the overlay starts the job instantly, skipping the app-launching dance.
What you probably won't use it for: long-form rewriting or deep research. Phi Silica is intentionally small—it prioritizes speed and privacy over raw power. For lengthy documents or nuanced creative writing, you'll still want Copilot in the cloud or a full-fledged LLM. And OCR accuracy, while good on clean fonts, stumbles on stylized text, low-contrast elements, or composite web pages. In those cases, old-fashioned copy‑and‑paste often feels faster.
The Hardware Hurdle: Why Your CPU Decides What You Get
Here's the catch. The complete Click To Do experience—especially the on-device text intelligence—requires a Copilot+ PC with a neural processing unit (NPU) that hits roughly 40 TOPS (trillions of operations per second). Devices with Qualcomm Snapdragon X series, Intel Core Ultra with integrated NPU, and AMD Ryzen AI processors qualify, but rollout has been staggered. Many Intel and AMD machines are still catching up on drivers and firmware, so the full feature set may not be available on day one for all Copilot+ labeled PCs.
If you're on a standard Windows 11 machine without an NPU, you may still see basic Click To Do actions—screenshot capture, URL detection, email launching—but the local AI transformations won't appear. That creates a two-tiered experience: buyers of premium laptops get a frictionless AI assistant layered into the OS; everyone else gets a glorified screenshot tool.
Microsoft has indicated that wider hardware support is coming, but for now, the dividing line is real. Before you upgrade or buy a new laptop specifically for Click To Do, confirm your exact model meets the 40 TOPS NPU spec, and check that Insider builds (or stable channel updates) have actually enabled the feature on your silicon.
Privacy: What Stays and What Leaves Your PC
Click To Do’s privacy architecture is a mixed bag of genuinely smart design and lingering trust questions.
On the positive side: the feature is opt‑in by invocation—it only captures your screen when you explicitly trigger it. Phi Silica runs locally, so text summaries and rewrites never touch Microsoft's servers. Temporary files may be created in your local temp folder during processing, but Microsoft says they are not stored persistently. And cloud‑based actions are clearly labeled and require your explicit click; nothing is sent silently in the background.
But privacy critics point to three pressure points. First, the screen‑capture mechanism itself, even when user‑initiated, can feel invasive—especially given Windows Recall’s controversial rollout history. Second, while the menu distinguishes local from cloud actions, the distinction isn't always obvious to a casual user. Selecting "Ask Copilot" might seem like just another text option, but it beams your content to Microsoft's cloud. Third, Microsoft's diagnostic telemetry around Click To Do isn't fully detailed in consumer-facing docs; enterprise admins will need to audit what data is collected and how it aligns with compliance policies.
For organizations, the calculus gets more complex. Cloud actions on managed devices route through Microsoft 365 Copilot, which respects organizational privacy controls, but that only works if the user is signed into a work account and IT has configured policies correctly. Home users on free Copilot accounts will have their data handled according to Microsoft’s standard consumer privacy statement.
Steps to Start Using (or Disabling) Click To Do
Ready to try it? Here's what to do.
- Check your hardware. You need a Copilot+ PC with an NPU. Look for the Copilot+ PC branding in Settings > System > About, or check your specific model against Microsoft’s compatibility list.
- Ensure you're on a compatible build. Click To Do is rolling out to Windows 11 Insiders first (Dev and Beta channels) and then gradually to stable builds. Go to Settings > Windows Update and check for updates.
- Invoke it. Press and hold the Windows key, then click the left mouse button. You can also use Windows+Q, or the Snipping Tool shortcut (Print Screen key if configured).
- Select your target. Hover over a text block or image area to highlight it. If the selection contains more than 10 words, the AI text actions will appear.
- Choose an action. For local privacy, stick to Summarize, Rewrite, or Create bulleted list. For web results, pick Visual Search or Search the web—but know your content is leaving the device.
If you find the overlay intrusive—accidental triggers during gaming are a common complaint—you can disable it entirely. Go to Settings > Privacy & security > Click to Do and toggle the feature off. This stops the shortcuts from working and hides the overlay.
For best results, work with high-contrast, legible text to improve OCR accuracy. If a local rewrite feels too narrow, consider sending the text to Copilot’s cloud chat for a more robust edit. And if you're on an enterprise machine, check with your IT department about which actions are allowed under your organization’s data governance rules.
Why This Matters Beyond Convenience
Click To Do is more than a handy utility; it's a tactical move in Microsoft’s dual AI strategy. By pushing on‑device models like Phi Silica for small, fast tasks, Microsoft reduces latency and addresses privacy concerns that have dogged Copilot features. The NPU in Copilot+ PCs becomes a selling point, not just a spec sheet checkbox. And by bundling scattered AI tools—image edits, text summaries, visual search—into one universal overlay, Microsoft is making AI a first‑class OS element, like the Start menu or taskbar.
Yet the approach also exposes the messy transition period we're in. Local models are useful for micro‑tasks but lack the depth of cloud rivals. Hardware gating means only a fraction of users can access the full experience right now. And the privacy messaging, while clearer than with Recall, still demands vigilance from users and admins alike.
Outlook: What to Watch Next
Click To Do will evolve in step with Windows’ AI roadmap. Microsoft is expected to expand NPU support to all Copilot+ certified devices in the coming months, closing the hardware gap. Updates to Phi Silica could bring longer context windows and more capable local reasoning. And tighter integration with third‑party apps—imagine a "Send to Photoshop" action—might turn the overlay into a genuine platform.
For now, Click To Do is a solid productivity booster for those with the right hardware, and a preview of a local‑first AI future for everyone else. Its success will depend on whether Microsoft can deliver on broader compatibility, improve local model quality, and maintain the privacy transparency that skeptics demand.