Microsoft Copilot

Microsoft Copilot is in the middle of its biggest product-identity change since Bing Chat became Copilot. Starting in August 2026, Microsoft began combining the consumer Copilot app and the Microsoft 365 Copilot app into one unified **Copilot** experience across web, mobile, Windows and macOS.¹²³

The unification simplifies the interface, but it does **not** turn every Copilot into the same product. Personal Microsoft accounts and work/school accounts remain separate, organizational security and compliance controls continue to apply to work data, and paid Microsoft 365 plans unlock different productivity and agentic capabilities.¹⁴

That distinction is the central fact in reviewing Copilot correctly. “Microsoft Copilot” can now refer to one app that contains a free personal AI companion, Microsoft 365-connected home features, and enterprise Microsoft 365 Copilot experiences depending on the account. Copilot Studio, Security Copilot and GitHub Copilot remain separate products despite sharing the name.

This is a desk review. AiToolMap did not perform or simulate its own hands-on test.

Microsoft’s current support documentation says Copilot is being updated into a more cohesive experience across personal, work and school contexts.¹ Windows Central and The Verge independently confirm the staged rollout beginning in mid-August 2026, first across mobile and web and later across Windows and macOS.³²

Personal and work accounts remain isolated. Work or school data does not become available to a personal account simply because the two experiences now share the Copilot app. Organizational security, privacy, compliance and administrative controls continue to apply when Copilot is used with an organizational account.²⁴

This is the right architecture, but it creates a review problem. The name on the icon no longer tells you exactly which data protections, model routing or capabilities apply. A user can switch between personal and work contexts inside one branded environment while those contexts remain legally and technically different.

Microsoft is also removing features during the transition. Group Chat threads and shared content do not carry forward after the account update; Podcasts was retired after August 18; and consumer Deep Research is being retired, with Researcher positioned as the advanced-research route for Microsoft 365 Premium subscribers.¹

That is substantial product churn, not a cosmetic redesign.

The strongest current fixed-panel hands-on test comes from Windows Central. The reviewer spent a month using the free versions of Copilot and Gemini to proofread real writing.¹⁸

Copilot was more conversational and personal. It was particularly useful for understanding the flow of a document and suggesting rearrangements or phrasing changes. Gemini was often more thorough and accurate on mechanical corrections and spent less time on conversational pleasantries. Both were substantially more context-aware than the old Microsoft Editor.¹⁸

That is a narrow task test, but it captures a broader product choice. Copilot has increasingly been designed as a **companion**, not just an answer engine. That can make interactions feel more natural, but personality is not the same thing as precision. For editing or factual work, a warmer response is only useful if the underlying correction is right.

Google Play reviews add a much larger but less controlled signal. The exact Android app carries a high aggregate rating across millions of reviews, while visible user comments include both praise for conversational usefulness and complaints about feature changes or usability regressions.¹¹

YouGov provides another lens. In current UK brand tracking, Copilot’s net Satisfaction is 37.2 among current/former users. Awareness is high, but preference and satisfaction are not category-leading.¹⁹ That is consistent with a product that has exceptional distribution without necessarily being every user’s preferred assistant.

Copilot is no longer confined to generic productivity. Copilot Health is a dedicated US preview for adult Microsoft 365 Personal, Family and Premium subscribers. Users can discuss symptoms, labs and care questions, and optionally connect health records or wearable data.¹⁰

Microsoft explicitly says Copilot Health is not intended to diagnose or treat conditions and is not a substitute for professional medical advice. Health conversations are stored separately from normal Copilot conversations, Health data is not used to train AI models or create advertising profiles, and users can delete their Health data.¹⁰

The default retention for health conversation activity is 18 months. Microsoft says connected wearable or electronic-health-record data is accessed in context rather than stored directly in Copilot Health, while a Health Profile can retain information the user asks Copilot to remember.¹⁰

Axios’s analysis of more than 500,000 Copilot health and well-being conversations shows why the safety design matters: users already ask substantial numbers of symptom, treatment and care-navigation questions, with mobile users tending toward more urgent and personal queries.²⁰

Bloomberg likewise documents Microsoft’s expansion into persistent personal-health context and the broader attempt to make Copilot feel like a durable personal companion.²¹

Health is therefore both a product opportunity and a test of trust. The dedicated no-training/no-advertising rules are materially stronger than the ordinary consumer Copilot regime.

The current consumer baseline is free Copilot. Microsoft does not require a €22 monthly Copilot subscription just to use the general assistant.⁷

In Italy, current individual Microsoft 365 pricing is €10/month or €99/year for Personal, €13/month or €129/year for Family, and €22/month or €219/year for Premium.⁷

Personal adds desktop Office apps with Copilot and higher use of selected features than the free tier. Family extends the broader Microsoft 365 bundle to up to six people, while the AI entitlement is tied to the subscription holder. Premium adds higher Copilot usage, advanced capabilities and AI agents for complex work such as research reports and data analysis.⁷

This bundle structure is important for value comparisons. If a user already pays for Microsoft 365 for Office and OneDrive, Copilot’s incremental value can be substantial. If the user only wants a standalone frontier chatbot, Microsoft 365 Premium is competing with dedicated AI subscriptions that may provide a clearer model/capability entitlement.

The retirement of consumer Deep Research also makes the distinction sharper: sophisticated Microsoft research workflows are moving toward Premium/Researcher rather than remaining a universal free Copilot capability.¹

Copilot should not be scored by taking the best available GPT or Microsoft model benchmark and pretending it describes every response in the product.

Computerworld reports Microsoft is increasingly designing Copilot around a **multi-model** architecture: memory, context and orchestration are separated from any single foundation model, allowing Microsoft to use its own MAI models and third-party models according to task and product surface.¹⁵

TechCrunch reports GPT-5.6 becoming a preferred model in **Microsoft 365 Copilot**, while Microsoft simultaneously develops and deploys its own models. That is useful evidence about the enterprise sibling, not proof that every free consumer Copilot request runs on GPT-5.6.¹⁶

Artificial Analysis’s current general-work-agent comparison similarly treats Copilot primarily as an integrated agent ecosystem spanning Microsoft 365, Microsoft Graph and connectors, not as a clean single-model benchmark target.¹⁷

This is why AiToolMap does not import ARC, LM Arena, LiveBench, HELM, Aider or SWE-bench scores into the consumer Copilot rating. The product routing is not sufficiently fixed and disclosed for that transfer to be defensible.

The absence of a clean model score is not the same thing as weak AI. It simply means Copilot should be judged more at the **product and workflow level** than by a foundation-model leaderboard.

The unified Copilot branding should not be read as a unified privacy policy.

For the updated post-August-18 consumer app, Microsoft’s privacy documentation distinguishes personal Microsoft accounts from work/school organizational accounts and from Copilot used inside Microsoft 365 applications.⁴

For eligible consumer Copilot use, Microsoft says conversation activity can be used to train generative AI models unless the user opts out. Users can disable future conversation and voice activity from model training while retaining personalization if they choose.⁸

Microsoft says it excludes organizational Entra users, Copilot conversations inside Microsoft 365 consumer apps, minors and certain jurisdictions from this consumer training regime. It also says identifying information is removed from training data and that personal account data such as profile details or email contents is not used for model training.⁸

That is more nuanced than either “Microsoft trains on everything” or “Copilot never trains on your data.” The correct answer depends on **where and how** the user is using Copilot.

The practical lesson is simple: a personal Copilot chat is not automatically equivalent to an enterprise-protected Microsoft 365 Copilot session. Sensitive corporate work should be performed in the account/product context whose data-protection terms actually cover it.

Several current sources document security failures in Microsoft 365 Copilot. TechCrunch reports a bug in which Microsoft 365 Copilot processed confidential labeled email despite data-loss-prevention controls, with Microsoft confirming and fixing the issue.¹⁶

VentureBeat separately reports enterprise sensitivity-label and DLP trust-boundary failures involving Microsoft 365 Copilot.²³

These incidents matter because Microsoft’s main competitive advantage is access to organizational context. The more useful an assistant becomes because it can see company email, files and collaboration data, the more important permission and policy enforcement becomes.

But this is enterprise **Microsoft 365 Copilot** evidence. AiToolMap does not turn these incidents into claims that the free consumer app leaked personal chats. Surface separation matters in both directions: enterprise integration strengths cannot inflate the consumer score, and enterprise security incidents cannot automatically depress it.

Copilot is an easy recommendation to **try** because the core assistant is free. It is especially sensible for Windows users and people already paying for Microsoft 365.

Microsoft 365 Personal or Family becomes more attractive when the user values Office integration and storage as much as AI. Premium makes sense for heavier users who specifically want advanced agentic research/analysis and higher Copilot allowances.⁷

Users who choose an AI product mainly by foundation-model benchmark rank may find Copilot frustrating to compare because routing is intentionally multi-model and not cleanly exposed. Users who care more about workflow integration than model identity may see that as a strength.

For organizational work, evaluate Microsoft 365 Copilot under enterprise data-protection terms rather than assuming the consumer app’s privacy rules apply. Conversely, do not assume enterprise integrations are included simply because the unified app carries the same Copilot icon.