Microsoft has quietly equipped its Copilot assistant with the ability to reason across multiple files at once, bringing a ChatGPT-like multi-document analysis workflow to the web and Windows 11 app for free. The update, first reported by Windows Latest and confirmed by a Microsoft representative, allows Copilot to ingest several documents and synthesize information across them — connecting dots rather than answering about each file in isolation.

Previously, Copilot could accept up to 20 file uploads in a single message, but it processed each file independently. Users could ask about one document at a time, but the assistant couldn't compare job descriptions against a résumé, generate a unified study guide from multiple lecture notes, or cross-reference a budget spreadsheet with an itinerary PDF. That changes now.

How the multi-file feature works

When a user uploads two or three files — the confirmed limit for the initial consumer rollout is three, though Microsoft's OneDrive Copilot supports up to five — Copilot's underlying engine, now powered by GPT-5 via a smart model router, detects the relationships among the documents and performs cross-document analysis. The router decides whether to use a lightweight model for simple queries or escalate to a deeper reasoning variant for complex synthesis, balancing latency, cost, and quality.

In practice, a hiring manager can upload two job postings and a candidate's CV, then ask, "Which role fits this candidate best and why?" Copilot will compare skills, experiences, and requirements across all three files, delivering a ranked recommendation with evidence drawn from each document. A student can hand it three lecture PDFs and ask for a practice quiz; Copilot will pull concepts from all three, generate questions, score answers, and offer explanations — all within the chat interface.

Windows Latest demonstrated exactly that during its hands-on test. Using the new "Study and Learn" mode (accessible from the "+" menu in the Copilot interface), the reporter uploaded three study documents and asked for a flashcard-based quiz. Copilot produced a set of questions, allowed inline answer selection, and provided a score with detailed feedback — behavior that closely mimics ChatGPT's Study Mode, now freely available inside Microsoft's ecosystem.

The GPT‑5 backbone and broader Copilot evolution

This multi-file capability didn't arrive in a vacuum. On August 7, 2025, Microsoft announced the integration of GPT‑5 across Copilot surfaces, complete with a smart routing system that dynamically chooses the appropriate model for each prompt. The company also introduced conversation modes: Quick for fast responses, Think Deeper for more thorough reasoning, and Deep Research (a paid/subscriber feature) for long-running, citation-backed analysis across multiple sources.

Those modes directly enhance multi-file work. A user can toggle Think Deeper while comparing three contracts, prompting Copilot to spend more compute cycles identifying subtle discrepancies. Deep Research can be invoked for substantial projects like combining a business plan, market research PDFs, and financial spreadsheets into a comprehensive investment memo — a workflow that previously required ChatGPT's Advanced Data Analysis or manual collation.

Microsoft has also been expanding file-aware tools in OneDrive, where Copilot already allows comparing and summarizing up to five files. The consumer web and Windows 11 app experience now inherits a version of that capability, simplified for everyday use. The overarching strategy, according to Microsoft's blog posts and release notes, is to make Copilot a first-class reasoning layer across Windows, Microsoft 365, and the web — with multi-file synthesis as a cornerstone.

Practical use cases beyond study aids

While the initial demos lean educational, the feature's real-world potential spans several domains:

  • Recruiting and HR: Upload multiple job descriptions and a résumé; ask for a fit score, skill gap analysis, and suggested interview questions.
  • Travel planning: Combine a budget spreadsheet, a drafted itinerary PDF, and a packing list; ask Copilot to identify cost-saving opportunities, highlight scheduling conflicts, and recommend local attractions.
  • Legal and contract review: Feed in two versions of a contract plus an amendment and receive a single annotated summary of changes, potential risk clauses, and action items.
  • Event organization: Merge venue proposals, catering menus, and a guest list to produce a consolidated planning document with timeline and budget alerts.
  • Research and reporting: Upload several whitepapers or market reports; get a unified literature review, trends summary, and citation index.

In each case, the assistant moves from Q&A bot to collaborative analyst — a shift that dramatically reduces the manual work of opening, reading, and cross-referencing multiple files.

Limitations to keep in mind

The Windows Latest report specifies that Microsoft confirmed a three-file reasoning limit for the consumer chat surface. This cap is not yet published in Microsoft's official documentation, so it may vary by region, account type, or as the feature scales. OneDrive Copilot, for comparison, supports up to five files for certain operations. Users should test with their own content to understand actual behavior.

File type and formatting also affect results. Text-based PDFs and Word documents work smoothly, while scanned PDFs rely on OCR pipelines that may introduce errors. Spreadsheets can be parsed, but complex formulas or multi-sheet workbooks may be flattened — a known limitation inherited from the underlying model's data handling. Images are processed through vision models, so handwritten notes or photos of bookshelves (as suggested by Windows Latest) will be analyzed, but output fidelity depends on image quality.

Above all, multi-file synthesis remains an AI-powered draft. Hallucinations — confident but incorrect cross-document connections — are a real risk, especially when sources contain conflicting information. Microsoft's own guidance warns users to verify any content used for legal, financial, or safety-critical decisions. A good practice is to prompt Copilot to cite the source document and line for every key assertion, then fact-check those references.

Under the hood: chunking, retrieval, and model routing

From a technical standpoint, Copilot's multi-file reasoning likely follows a hybrid architecture common to modern AI assistants:

  1. Chunking and indexing: Each file is split into segments, converted to embeddings, and stored in a session-specific index. When a query arrives, the system retrieves the most relevant chunks from all files.
  2. Context stitching: Retrieved chunks are assembled into a prompt that fits the model's context window, prioritizing the most salient passages. This allows the model to "see" related information from different files side by side.
  3. Model routing: Copilot's smart router decides whether the assembled context and query complexity warrant a lightweight model (for simple lookups) or a heavier GPT‑5 reasoning variant (for synthesis). Multi-file tasks, particularly those requiring comparison or creative aggregation, are more likely to hit the reasoning tier.
  4. Specialized pipelines: For spreadsheets, a sandboxed kernel may parse tables and perform deterministic calculations before feeding results back into the LLM. For images, a vision pipeline extracts text or describes visual elements.

This pipeline explains both the feature's power and its constraints. The three-file cap likely exists to keep retrieval and reasoning costs manageable for the free tier, while Deep Research mode (which can handle more extensive multi-document work) is reserved for subscribers to cover higher compute expenses.

Privacy and enterprise governance checklist

Because the consumer Copilot processes uploads in Microsoft's cloud, organisations and individuals must treat the feature with appropriate caution:

  • Data sensitivity: Do not upload confidential, regulated, or personally identifiable information unless you are certain the processing environment meets your compliance needs. By default, consumer sessions are not isolated to enterprise tenants.
  • Enterprise alternatives: For governed workloads, use Microsoft 365 Copilot or Azure AI Foundry, which offer tenant-scoped controls, audit logs, and data residency commitments. Those services are designed for multi-file analysis with compliance in mind.
  • Output verification: Implement a human-in-the-loop review for any critical documents — contracts, regulatory filings, medical summaries. The assistant can accelerate drafting but must not be the final authority.
  • Rate limits: Microsoft does not publicly document precise per-user quotas for the free consumer multi-file feature. If you plan heavy daily usage, test the limits or consult your Microsoft account representative.

These precautions mirror the standard advice for any cloud-based AI processing, but the temptation to drop sensitive files into a free, convenient tool can be high — so governance must be explicit.

How to adopt multi-file Copilot today

If you want to incorporate this into your workflow, follow a phased approach:

  1. Start with three representative files of the type you'll regularly use (contracts, notes, spreadsheets). Observe whether Copilot treats them as a combined context or answers separately. If the router falls back to per-file responses, rephrase your prompt to explicitly ask for cross-document analysis.
  2. Choose the right mode. For quick comparisons, the default Quick mode suffices. For deeper, validated outputs, toggle Think Deeper. For subscription-tier projects, invoke Deep Research — expect longer turnaround but more structured, source-linked results.
  3. Feed explicit instructions. In your prompt, say “Treat these three files as a single corpus and extract the top five themes, citing sources.” This helps the router select the appropriate model.
  4. Verify systematically. After receiving an answer, ask Copilot to highlight which claim came from which file. If it can't, treat the output as unreliable.
  5. Scale carefully. If you plan to automate nightly reports by feeding multiple files, pilot the flow first. Measure hallucination rates and keep a human reviewer in the loop until confidence stabilizes.

Microsoft's addition of ChatGPT-style multi-file analysis to Copilot's free consumer surface is a practical upgrade that lowers the barrier to sophisticated AI reasoning. The combination of GPT‑5, smart routing, and focused modes like Study and Learn turns Copilot into a genuine productivity multiplier — not just a chat sidebar. But the feature's power comes with responsibility. Users and administrators must stay clear-eyed about its limitations, verify outputs, and choose governed paths for sensitive data.

For everyday tasks — comparing job offers, building study guides, planning trips — Copilot's new multi-file brain is a welcome evolution. It closes a key feature gap with ChatGPT and, for the first time, brings contextual, cross-document analysis to Windows and the web without a subscription. Just remember to check the work before you act on it.