Microsoft shipped a Windows 11 Insider preview build this week that finally makes the operating system’s search understand what you mean, not just what you type. The long-awaited semantic search feature can interpret natural-language phrases like “find the budget spreadsheet from last quarter” or “photos of a dog at the beach,” but there’s a significant catch: it only works on Copilot+ PCs equipped with a neural processing unit (NPU).
What’s New: Semantic Indexing Comes to Windows
The feature arrives in Windows 11 Insider Preview Build 26120.2992 (KB5050083) and marks the first time Microsoft has shipped a meaning-based search layer inside the OS. Unlike the traditional lexical index—which matches exact filenames, metadata, or text strings—the new semantic index analyzes content and context to understand intent. Microsoft combines both layers in a hybrid approach: the legacy index handles fast, precise lookups, while the semantic layer ranks and returns results that match your description even when the exact keywords aren’t in the file name.
Early hands-on testing, including Thurrott’s detailed walkthrough, shows the system handling queries like “Europe trip budget” and returning spreadsheets, Word documents, and PDFs about that topic—regardless of file names. The semantic capabilities extend beyond text: optical character recognition (OCR) lets you search for “photos of a bridge at sunset” and get results from your image library.
Crucially, all inference runs locally on the device using the NPU in Copilot+ PCs. Microsoft says the processing stays on your machine, with no cloud round-trips, which both preserves privacy and works offline. Supported file formats in the preview include common document and image types (.docx, .pdf, .xlsx, .png, .jpeg, .txt, .csv, etc.) and the feature currently supports English, Chinese, French, German, Japanese, and Spanish, with more languages planned.
Surfaces that tap into semantic search include:
- The taskbar search box
- File Explorer (with some cloud photo surfaces)
- Windows Settings (natural-language queries to find preferences)
- The Copilot app, which gains a redesigned homepage and permission controls
Who Can Actually Use It Today
If you don’t have a Copilot+ PC, you’re out of luck for now. Microsoft has gated the preview to devices with that certification, which today means Snapdragon X-series powered laptops. Intel and AMD Copilot+ hardware will follow, but no timeline has been announced. Even on eligible hardware, the feature is rolling out slowly; not every Insider with a Copilot+ PC will see it immediately after updating to build 26120.2992.
The hardware requirement stems from the computational demands of running semantic models locally. Copilot+ devices ship with NPUs capable of 40+ TOPS (trillions of operations per second), which Microsoft deems necessary for low-latency, power-efficient on-device AI. Running the same work on a CPU or GPU would either be too slow or force a cloud dependency—defeating the privacy and offline benefits.
What Semantic Search Means for Your Daily Workflow
For home users and students
If you own a Copilot+ laptop, the upgrade is immediate and useful. You can stop memorizing exact file names and folder paths. A query like “find my resume draft” works even if the file is named “document_final_v3.docx.” The integration with Copilot lets you jump from finding a file to taking an action—summarizing a PDF, identifying an image—without switching apps. Just remember that search covers only indexed locations; if your important folders aren’t included, you won’t see results from them.
For power users and enthusiasts
The hybrid index architecture means you can still fall back to traditional exact-match searches when you need precision, while the semantic layer handles ambiguous or conversational queries. The offline, on-device processing is a genuine privacy win, but you’ll want to audit which directories are indexed—enable “Enhanced Indexing” in settings if you need broad coverage, but be cautious about sensitive folders. The Copilot app’s permissions page lets you restrict which files it can touch; treat it like any app with file system access.
For IT administrators and enterprises
This is a pilot-only technology right now. Microsoft has not provided detailed documentation on telemetry, cache retention, or admin controls. The promise of local-only processing reduces cloud risk, but until Group Policy or Intune controls emerge, you can’t enforce search behavior across a fleet. For regulated industries, validate whether semantic index artifacts are persisted, stored, and if any metadata ever leaves the device. Run a small test group on Copilot+ hardware, demand written documentation from Microsoft, and educate users about the difference between local search and explicit file uploads to Copilot.
For developers
The underlying semantic indexing technology is not yet exposed as a public API, but it signals Microsoft’s direction: meaning-first search will eventually permeate the OS. If you build Windows applications that manage files, consider how your users might expect natural-language retrieval in future Windows releases.
How We Finally Got Here: The Long Road to Meaningful Search
Windows Search has frustrated users for decades. The core indexing engine—introduced in Windows Vista—relied on exact matches and metadata, leaving users to guess filenames or wrestle with advanced query syntax. Cortana promised a conversational interface but never deeply understood file contents. Microsoft added basic content indexing over the years, but true semantic search remained elusive.
The breakthrough came from the convergence of on-device AI hardware and vector embedding technology. Microsoft’s Copilot+ certification, launched in 2024, defined a new class of PC with NPUs capable of 40+ TOPS. That hardware headroom made it practical to run semantic models locally without draining battery or slowing the machine. Build 26120.2992 is the first to wire that capability into Windows search surfaces, turning an Insider preview into a real-world test bed.
Your Action Plan: What to Do Right Now
Check your hardware. Open Settings > System > About and look for “Copilot+” branding, or run “msinfo32” to confirm your processor. Currently, only Snapdragon X Elite and X Plus systems qualify.
Join the Insider program. Go to Settings > Windows Update > Windows Insider Program and enroll in the Beta or Dev channel, then check for updates to get build 26120.2992. After installation, search for a common query—“documents about budget”—to see if the feature has been enabled for your device.
Configure indexing wisely. Navigate to Settings > Privacy & security > Searching Windows. Turn on “Enhanced” indexing if you want semantic search across your entire PC, but first review what’s in your user folders. Consider moving sensitive files to a non-indexed location or encrypting them separately.
Lock down Copilot app permissions. Open the Copilot app, go to its settings, and review which folders it can access. Uncheck any directories you don’t want Copilot to see, even for local-only operations.
Pilot before rolling out in business. If you manage devices, buy or designate a few Copilot+ machines for a controlled test. Document how search behaves, check for any unexpected network traffic during heavy indexing, and press your Microsoft representative for a whitepaper on governance and telemetry.
If you’re on x86 hardware, wait. Intel and AMD will eventually ship Copilot+ certified systems, and Microsoft says support will expand. Until then, you can’t run semantic search locally. Don’t invest in new hardware solely for this feature unless you also value other on-device AI capabilities.
Looking Ahead: What to Expect Next
Semantic search is unlikely to stay behind a Copilot+ wall forever. As NPUs become more common, the feature will spread to a broader range of devices. In the near term, expect Microsoft to add Intel and AMD Copilot+ support, more file formats, and additional languages. Enterprise controls will eventually surface through Group Policy and Intune, driven by feedback from early testers. Longer-term, semantic indexing may underpin smarter file recommendations, AI-driven workflows, and deeper integration with OneDrive. For now, it’s a limited but meaningful upgrade that finally answers the decades-old complaint: Windows search actually works—just not on your current PC.