Mobile AI assistant usage has decisively overtaken desktop, with Microsoft Copilot, Google Gemini, and OpenAI ChatGPT driving record growth on smartphones even as PC engagement slides. New data from Comscore’s AI usage tracker reveals that between March and June 2025, mobile reach for 117 tracked AI tools hit 73.4 million U.S. users—a 5.3% increase—while desktop usage fell 11.1%. From November 2024 to June 2025, mobile adoption of AI assistants surged 82% overall. The fastest-growing app on mobile? Microsoft Copilot, which rocketed up 175%, followed by Google Gemini at 68% and ChatGPT at a still-strong 17.9%.

These are not vanity metrics. They capture a fundamental shift in how people integrate AI into daily life: away from desk-bound queries and toward on-the-go, multimodal interactions that feel native to the phone in your pocket. Smriti Sharma, senior vice-president of custom IQ at Comscore, frames it as a behavioral sea change: “The evolution of how consumers use AI tools on mobile isn’t just about convenience, it’s about behaviour. Leading brands are positioning AI assistants as personal, always-available companions, and users are responding.”

For Windows users and IT administrators, this mobile pivot carries immediate—and sometimes uncomfortable—implications. Copilot’s triple-digit growth is not merely a consumer trend; it is being fed by enterprise licensing, OS-level bundling, and admin-driven rollouts that turn the assistant into a default across devices. The same forces that accelerate adoption also raise hard questions about governance, data privacy, and platform lock-in.

The data behind the shift

Comscore’s tracker, launched in May 2025, monitors deduplicated audience reach across mobile and desktop for 117 AI tools in nine categories—from general-purpose chatbots to coding assistants and image generators. Using a representative digital panel that captures both web visits and native app usage, it provides a rare apples-to-apples comparison of cross-platform behavior.

Key numbers from the March–June snapshot:

  • Total mobile reach (web + native apps): 73.4 million unique users, up 5.3%.
  • Desktop/PC reach: down approximately 11.1% in the same period.
  • Overall mobile AI assistant growth from November 2024 to June 2025: 82%.
  • Fastest mobile growth by brand: Microsoft Copilot +175%, Google Gemini +68%, OpenAI ChatGPT +17.9%.

Critically, these figures reflect deduplicated audience reach—the number of distinct individuals who used a tool on a device at least once—not raw session counts or API call volumes. That distinction matters when comparing stories from different trackers. Web referral analyses (such as those from StatCounter) have long shown ChatGPT commanding a dominant share of outbound traffic from chatbot sites. Comscore’s panel-based view instead answers the question “How many unique people are using these tools on their phones or PCs?”—a complementary but distinct lens.

Percentage growth figures demand scrutiny. A 175% jump for Copilot is dramatic, but it starts from a smaller absolute base. Secondary reports estimate that in this window Copilot’s mobile user base reached roughly 8.8 million, while ChatGPT’s stood at around 25.4 million—meaning Copilot’s growth, while explosive, still trails the scale incumbent. Without published absolute counts from the vendors themselves, these secondary estimates should be treated as directional benchmarks, not hard facts. The lesson: never read a growth headline without asking “From what base?”

Why phones are winning casual AI interactions

The mobile takeover isn’t surprising to anyone who has dictated a text, photographed a receipt for a travel report, or asked an assistant to summarize a meeting note while walking between conference rooms. Mobile AI sessions tend to be short, context-specific, and multimodal. Users want quick answers, not sprawling conversations. Voice input, camera capture, and tight OS integration dramatically lower the friction compared with opening a browser tab on a laptop.

Performance realities reinforce the trend. Mobile networks and on-device processing demand low latency and efficient models. Assistants that deliver snappy, relevant responses win repeat usage. As a result, vendors are steering product roadmaps toward smaller, optimized models and hybrid inference strategies that split work between the device and the cloud. UX polish—not raw parameter counts—is becoming the mobile differentiator.

For Microsoft, the mobile surge opens a path to more deeply embed Copilot into the Windows ecosystem. A user who drafts an email with Copilot on their phone during a commute can later refine it on a Windows PC, with context seamlessly carried over. This cross-device stickiness is precisely what enterprise buyers should be factoring into licensing and rollout decisions.

Platform-by-platform breakdown

Microsoft Copilot: The enterprise wedge driving mobile lift

Copilot’s 175% mobile growth did not happen solely because consumers discovered it in the App Store. The assistant benefits from multiple low-friction distribution channels: it lives inside Microsoft 365 apps, the Edge browser, Windows, and enterprise licensing agreements that let IT organizations provision Copilot across managed devices—including personal phones enrolled in BYOD policies. Single sign-on, admin controls, and existing productivity investments create a scenario where Copilot is enabled rather than chosen.

For enterprises already standardized on Microsoft 365, this bundling is a double-edged sword. It promises rapid productivity gains and simpler management, but it also reduces user agency and raises questions about vendor lock-in. IT teams must recognize that Copilot’s mobile growth is as much a function of distribution architecture as of product quality.

Google Gemini: Default placement and Android leverage

Google’s 68% mobile growth rides on device preloads and deep Android integration. Pixel phones ship with Gemini as the default assistant, and even non-Pixel Android devices increasingly surface Gemini features in Google apps and Workspace. Unlike Microsoft’s enterprise-led push, Google’s strategy leans on consumer ecosystem defaults—a time-tested path to massive scale. Long-context reasoning and multimodal features (such as analyzing a document while chatting) align naturally with Android’s strengths, making Gemini a strong contender for users who live inside Google’s app universe.

OpenAI ChatGPT: The loyalty leader at scale

ChatGPT’s 17.9% mobile growth may look modest beside the triple-digit surges of its rivals, but it reflects a product already at massive scale. What ChatGPT may lack in immediate percentage growth, it makes up for in stickiness. Comscore’s cross-visitation analysis shows OpenAI mobile users are the most loyal: fewer hop between assistants compared with users of Google or Microsoft tools. More than 85% of heavy AI assistant users across all platforms primarily stick to a single brand, and ChatGPT’s base holds particularly tight.

This loyalty creates a competitive moat. Developers integrate ChatGPT into their own apps, plugins multiply, and content partnerships deepen—all of which reinforce habit. For enterprises weighing multi-assistant strategies, ChatGPT’s retention power means a decision to allow it will likely lead to sustained, hard-to-displace usage.

Distribution as a moat

The divergent growth numbers are not just about product features. They reflect three distinct distribution engines:

  • Microsoft uses enterprise licensing and embedded productivity integrations to push Copilot into mobile handsets alongside desktop and web.
  • Google leverages device preloads, Android defaults, and Workspace cross-sell to make Gemini the path of least resistance.
  • OpenAI relies on brand mindshare, developer ecosystem, and direct-to-consumer product experience—a more traditional software adoption funnel.

Device preloads and enterprise enablement can produce rapid user-count acceleration, but they also shift the nature of consumer consent. When an assistant becomes the default on a corporate device or a new phone, the line between user choice and organizational mandate blurs. That carries significant governance implications, especially for regulated industries.

Loyalty, cross-visitation, and habit formation

Comscore’s deduplication engine reveals that more than 85% of top AI assistant users primarily use a single platform. AI assistants are no longer novelty toys that people sample and discard; they are becoming habitual workflow tools. Habit formation favors three things:

  1. Assistants that are defaulted by the OS or IT department.
  2. Deep integrations into frequently used apps.
  3. Personalization that makes results feel tailored and immediately useful.

Casual users still explore multiple tools, especially when new multimodal features—live camera analysis, voice-only modes, or image-to-text capabilities—appear uniquely on one platform. But power users quickly settle. For enterprises, this means the assistant chosen for broad deployment could become deeply entrenched within weeks, making switching costs high. Procurement teams should negotiate portability terms and audit trail access now, not after the rollout.

Caveats, governance, and the risk of lock-in

The mobile AI story is not without its shadows. Consider these practical concerns:

Measurement noise. Headlines built on percentage growth hide absolute scale. Different tracking methods—panel reach, referral share, API inference counts—tell different parts of the story. Decision-makers must triangulate across trackers and demand transparent definitions of “user,” “session,” and “visit” from vendors. Unverified cumulative user claims (some reaching into the billions) lack independent corroboration and should be flagged as unverified.

Privacy and permissions. Mobile assistants often require camera, microphone, location, and sometimes screen-reading permissions to deliver rich multimodal experiences. Each permission expands the attack surface for data misuse. For enterprises handling sensitive data, this demands clear consent flows, contractual data-usage boundaries, and DLP controls that prevent corporate documents from inadvertently being passed to third-party AI services through an assistant prompt.

Vendor lock-in. The habit formation advantages enjoyed by Microsoft and Google can morph into lock-in. An organization that standardizes on Copilot across its productivity stack may find it cumbersome to migrate later if pricing models change or if a competing assistant offers better features for specific tasks. Procurement teams should demand data portability, auditability, and dual-footprint flexibility in contracts.

Unverified hyperbolic claims. The wider market is awash with user-number rankings that lack transparent methodology. Viral lists often conflate cumulative downloads with monthly active users or include ambiguous “user” definitions. When evaluating competitive positions, rely on independently sampled data like Comscore’s panel plus multiple verification sources.

What this means for Windows users and IT admins

For Windows-centric organizations, Copilot’s mobile explosion is a strategic signal:

  • Leverage existing investments. If the organization already uses Microsoft 365, enabling Copilot across mobile devices can accelerate productivity gains without additional licensing complexity.
  • Govern first, deploy second. Before flipping the switch, establish data loss prevention rules, retention policies, and role-based access controls. An audit trail for assistant queries that touch corporate data or identity systems is non-negotiable in regulated environments.
  • Pilot with measurement. Run a controlled trial that instruments latency, hallucination rates, and real productivity impact. Don’t assume the marketing promise matches the user reality.
  • Plan for multi-assistant reality. Even if Copilot is the default, users will likely experiment with Gemini and ChatGPT. Build a governance framework that allows vetted, auditable models for sensitive workflows while permitting consumer-grade assistants for low-risk tasks.

For everyday Windows users, the takeaway is simpler: expect more AI features to appear across the apps you use, and treat permission dialogs with care. Be skeptical about uploading sensitive files or account credentials into any assistant prompt, even from a trusted vendor. Use paid tiers or enterprise configurations when data residency and stronger service-level agreements matter.

How to read the Comscore window

Comscore’s dataset is one high-quality lens—particularly strong for mobile and app behavior—but not the only one. Complement it with referral/session trackers (which measure outbound traffic shares) and vendor financial disclosures. The panel answers “How many unique people used this on a device?” while referral data answers “Which assistant sends the most traffic across the web?” Both are needed for a complete picture. When scale matters for procurement or infrastructure planning, insist on absolute figures and transparent methodology.

Looking ahead

The mobile AI shift is accelerating, and several trends will define the next phase:

  • Hybrid inference will become standard as vendors fuse on-device processing with cloud intelligence to shave latency.
  • Multimodal capabilities—camera, voice, text—will be baseline expectations, not differentiators.
  • Distribution deals (preloads, carrier bundles, enterprise enablement) will remain the fastest path to user growth, drawing increasing antitrust and procurement scrutiny.
  • Measurement must mature. The industry urgently needs standardized user metrics, independent audits of vendor-reported figures, and privacy-preserving telemetry frameworks that let product teams track adoption without compromising user data.

For Windows enthusiasts and IT leaders, the Comscore snapshot is a clarion call. Mobile AI assistants are no longer a sidecar to the desktop experience; they are becoming the primary interface for many users. The brands that pair product excellence with distribution muscle and trustworthy governance will write the next chapter. The question for the rest of us is whether we shape our policies and habits to match—or let defaults decide.