Google’s Gemini emerged as the clearer day-to-day assistant in a hands‑on ZDNET comparison against Microsoft’s Copilot, winning four of seven real‑world tasks — but the headline hides important nuance: each tool still shines in different workflows, and reliability, grounding, and ecosystem fit remain the deciding factors for Windows users.
Background
The public debate over AI assistants has moved beyond raw model benchmarks into practical, task‑based comparisons. Vendors now bundle assistants into large ecosystems (Google Workspace and Microsoft 365), and real‑world usefulness depends on three things:
grounding (ability to use live web or tenant data correctly),
tooling (maps, image generation, file/document access, agent capabilities), and
safety/governance (how enterprise data and user privacy are handled). Recent hands‑on tests run by journalists and reviewers reflect those axes rather than pure model math.
Two vendor facts worth verifying up front: Microsoft publicly rolled GPT‑5 into Copilot (making GPT‑5 available inside Microsoft 365 Copilot and Copilot Chat), and Google released Gemini 3 as a major model family update that powers new multimodal and agentic capabilities. Microsoft’s product pages and blog posts confirm GPT‑5’s availability inside Copilot and describe a router that selects faster or deeper reasoning models based on task complexity. Google likewise announced Gemini 3 as a major upgrade to its Gemini lineup, promoting improved reasoning, multimodal features, and new “Deep Think”/Pro modes. These model headlines matter, but they don’t settle the practical question: which assistant helps you get everyday work done on a Windows PC? The ZDNET head‑to‑head that prompted this feature tried to answer exactly that, using seven tasks a typical user might ask a desktop assistant to perform.
What the ZDNET test did (quick recap)
The reviewer executed identical prompts in each assistant (Gemini via Google’s web/app interface; Copilot via Microsoft’s Copilot/Edge integration) for seven everyday tasks and judged results on accuracy, creativity, and follow‑through. The assignment set was deliberately pragmatic — travel planning, map drawing, research on Windows history, infographic/image generation, a personal finance decision, a PowerShell scripting task, and movie trivia. The outcomes:
- Gemini — winner in itinerary planning, map task, infographic creation, and one other scenario.
- Copilot — clear winner on the PowerShell script task.
- Ties — research and finance advice; movie trivia also tied for correctness.
Below I summarize each challenge, explain why each assistant behaved the way it did, and offer a practical verdict for Windows users.
Challenge-by-challenge breakdown
1. Put together a trip itinerary — Gemini: better routing and destination selection
Gemini produced a sensible multi‑city European Christmas‑market itinerary using direct train routes and practical timing. It handled constraints (two nights per stop, final stop Strasbourg, trains under four hours) and accepted refinements (adding Cologne) without losing coherence. Copilot’s initial reply stayed conservative, producing an Eastern‑France–only route and citing direct‑train constraints that were demonstrably incorrect. After follow‑ups, Copilot admitted alternate routing was valid, but the initial misfire cost confidence.
Why this happened
- Gemini benefits from strong web grounding and tight Maps/Travel integrations that make route and location facts easier to surface in chat.
- Copilot’s strength is enterprise grounding, not public‑web route recall; when the system errs it often reflects conservative defaults or failure to consult live route data.
Practical takeaway: for exploratory travel planning and quick, map‑aware itineraries, Gemini is faster and more likely to produce directly usable routes; use Copilot when you need itineraries tied into Outlook calendars, travel expense workflows, or tenant data.
2. Draw a map — Gemini: pragmatic link‑based solution; Copilot: inaccurate graphics
When asked to produce a bird’s‑eye map for a multi‑city loop, Gemini recognized its limitations as a text/AI model and provided a direct Google Maps link with pins for the requested cities. That’s the pragmatic answer: it refused to fabricate a turn‑by‑turn vector map it could not guarantee. Copilot attempted to render a stylized map and badly misplaced cities (Munich in Czechia, Stuttgart in northern Italy), then eventually conceded it couldn’t meet the accuracy requirements.
Why this happened
- Gemini’s ecosystem ties enable it to hand off to Maps or generate map links reliably.
- Copilot’s creative visualization layer (when it tries to draw) can be faulty; generative “stylized” maps are hard to get right without geospatial grounding.
Practical takeaway: when you need an actual map, get a generated link or native map embed (Gemini or Maps); don’t trust freeform map images from a chatbot for navigation.
3. Research Windows history — Tie: both competent but verify
Both assistants returned correct release and end‑of‑support dates for Windows versions since XP, and both summarized the differences in system requirements between XP and Windows 7. Gemini earned a small editorial edge for noting the Windows 8 → 8.1 support nuance, but the results were close enough that a cautious reporter would still fact‑check. This is exactly the class of task where
both tools can accelerate research but shouldn’t replace primary sources.
Practical tip: use AI to assemble a draft timeline and then verify each date against vendor pages or archived Microsoft documentation before publishing.
4. Create an infographic about passkeys — Gemini: clearer composition and faster iteration
Gemini produced an attractive, informative infographic to illustrate the passkey concept (thumbprint → key → browser lock). Copilot offered generic icons with poor layout and didn’t iterate to a satisfactory result after multiple attempts. The reviewer found Gemini’s outputs both faster and more useful with smaller refinement cycles.
Why this happened
- Gemini’s multimodal image generation and layout tools are designed for quick conceptual art.
- Copilot’s creative image routines (and availability across tiers) can still be hit or miss, particularly in free/entry tiers.
Practical takeaway: for quick conceptual art or infographics for editorial use, Gemini is currently the better time‑to‑usable‑asset choice.
5. Help make a financial decision (lease vs buy) — Tie: both good at basic financial counseling
For a routine finance question, both chatbots asked clarifying questions and reached similar recommendations (buying in the tested scenario). This is an example of a low‑risk, high‑signal use case: many AI models can competently walk through calculations and tradeoffs if the underlying math is straightforward and the assistant asks the right follow‑ups.
Practical guidance: treat these outputs as decision
support — not authoritative financial advice, and always validate numeric assumptions (interest rates, depreciation schedules) with a calculator or spreadsheet.
6. Create a PowerShell script to rename JPEGs using metadata — Copilot: superior code output
This was Copilot’s decisive win. Copilot delivered a native PowerShell solution that prompted for folder paths, read metadata without insisting on third‑party tools, proposed error handling for missing location tags, and suggested undoability measures. Gemini struggled: it initially pushed third‑party ExifTool, produced scripts that failed on edge cases, and required multiple retries to produce a robust working script. For coding and platform‑specific automation, Copilot’s direct integration with the developer ecosystem and Microsoft’s focus on Windows workflows gives it a measurable advantage.
Why this happened
- Copilot is tuned for Windows scripting, PowerShell idioms, and developer workflows (and now runs on GPT‑5 models inside Microsoft stacks).
- Gemini is strong at reasoning and multimodal tasks but occasionally defers to external utilities for platform‑specific scripting.
Practical takeaway:
use Copilot for Windows automation and PowerShell code drafts, but still lint, test, and validate before running scripts on production data.
7. Movie trivia — Tie: both hit the right answer
Both assistants identified Dianne Wiest’s role in Bullets Over Broadway and the “Don’t speak” line. Here the task is simple factual recall and both systems performed correctly. Gemini’s answer was terse; Copilot’s was more expansive — but either would settle a party bet.
Cross‑checks and vendor claims: verified or flagged?
- Claim: “Copilot uses GPT‑5 as its default LLM.” Microsoft published that GPT‑5 is available in Microsoft 365 Copilot and that Copilot selects models (fast vs deep reasoning) automatically; documentation and blogs from Microsoft confirm GPT‑5’s integration into Copilot. That endorsement verifies the general claim, but reality is nuanced: vendors often roll new models progressively and may route sessions to lighter or deeper variants based on workload. Microsoft’s own documentation says GPT‑5 is being deployed and that model routing will be used to pick the right variant for each prompt.
- Claim: “Gemini 3 is smarter, faster, and free to access.” Google announced Gemini 3 as a model family with Pro/Deep Think modes; availability and pricing depend on tier (some new capabilities are Pro/Ultra features). Gemini’s improvements in reasoning and multimodality are real, but access depends on the subscription tier and the specific model variant (Flash vs Pro vs Deep Think). Reporters note improved speed and reasoning in Gemini 3, but the free access assertion requires nuance: free tiers often expose Flash or lower‑capacity variants, while Pro/Ultra unlock higher‑capability modes.
Where vendor claims are unverifiable
- Specific performance numbers (latencies, benchmark deltas) and internal “router” heuristics are vendor statements that can be confirmed only by controlled benchmarks; treat promotional performance claims as vendor‑framed until independent benchmarks appear. Flag these claims as “vendor‑reported” when used in decision‑making.
Strengths, risks, and governance considerations
Strengths (what I observed across tests)
- Gemini: excels at web‑grounded tasks, mapping/route outputs, polished multimodal art, and quick creative iterations. Best when you need on‑the‑fly visuals or web‑linked factual results.
- Copilot: excels at Windows‑centric automation, tenant‑grounded tasks inside Microsoft 365, and coding/PowerShell tasks tied to Microsoft platforms. Best for enterprise workflows, Outlook/Teams integration, and Windows administration tasks.
Risks (what to watch for)
- Hallucinations: both assistants can confidently assert incorrect facts — the BBC and other evaluations found significant error rates on news summaries and factual queries. Never publish AI output without verification for factual content.
- Ecosystem lock‑in: the productivity advantage often comes with vendor lock‑in: Copilot’s Microsoft Graph integration is powerful for enterprise data workflows but binds you to Microsoft’s governance surfaces; Gemini’s Workspace ties offer similar lock‑in. Match tool to data lifecycle and compliance needs.
- Privacy and training: consumer tiers may still ingest prompts into training pipelines (or retain telemetry) depending on vendor contracts; for regulated data, use enterprise plans with non‑training guarantees and tenant grounding.
- Image/model IP issues: image generation remains legally contested territory; when using generated art commercially, confirm licensing and provenance controls.
Practical recommendations for Windows users
- If you live inside Microsoft 365 (Outlook, Teams, OneDrive, SharePoint) and need tenant‑aware assistants for drafting, summarizing mail, or automating Excel/PowerPoint tasks, Copilot is the pragmatic choice. Its governance tools and admin surfaces matter for corporate deployments.
- If you need fast web‑grounded research, maps/route integration, or quick creative assets (infographics, layouts), Gemini will usually be faster and more polished. Use Gemini for ideation and rapid mockups.
- For scripting, code generation, and Windows automation (PowerShell), prefer Copilot — but always review and test generated scripts in a sandbox first.
- Keep two assistants in rotation as a fallback strategy: one for research/citation tasks and another for productivity/automation. This pluralism hedges vendor outages and model biases.
How to use these tools responsibly (a short checklist)
- Always ask the assistant for its sources when a factual claim matters; if the assistant doesn’t provide links, verify externally.
- For automation scripts or code, require unit tests, code review, and an undo plan before running changes that mutate files.
- For regulated or sensitive data, use enterprise‑grade connectors and check vendor non‑training and data residency terms.
- Treat images and creative assets as drafts until cleared for legal reuse; prefer services that expose provenance metadata.
Final verdict
The ZDNET hands‑on showed that the headline “Gemini beats Copilot” is true for a specific set of consumer‑oriented, web‑grounded tasks — particularly travel planning, mapping, and creative infographic generation — while Copilot remains the stronger assistant for Windows‑native automation and developer workflows such as PowerShell scripting. Both assistants are useful, but their usefulness is contextual:
Gemini for exploration and creative tasks; Copilot for Windows and Microsoft 365 productivity.
The sensible user strategy for the next 12 months is practical pluralism: pilot both assistants for your three most common workflows, pick the one that integrates with your data and compliance needs, and always insist on human verification for factual outputs or code that touches production systems. The race between model families (GPT‑5 in Copilot vs Gemini 3’s Deep Think modes) will continue to shift the balance, but ecosystem fit and governance will remain the business‑critical decision factors long after raw model scores make the headlines.
Source: ZDNET
Gemini vs. Copilot: I tested the AI tools on 7 everyday tasks, and it wasn't even close