AI Assistants Go Cross-Platform: Mobile Surges, Desktop for Deep Work (Comscore Dec 2025)

Comscore’s new snapshot makes one thing clear: the AI‑assistant era is not niche anymore — it’s a cross‑platform consumer phenomenon with mobile suddenly in the fast lane, and desktop still hosting the heavy‑duty sessions that power productivity and research.

Background / Overview​

Comscore published a short, sharp press release on January 29, 2026, reporting that mobile visitation to leading AI assistant destinations reached 54.3 million unique visitors in December 2025, up +107% year‑over‑year, while desktop visitation hit 83.0 million unique visitors, up +18% YoY. The release lists per‑product reach on each device, showing OpenAI’s ChatGPT leading on both desktop and mobile, with notable double‑ and triple‑digit percentage gains for several rivals including Google’s Gemini, Microsoft Copilot, and Perplexity.
Those headline numbers were syndicated widely (for example via GlobeNewswire), and they reflect Comscore’s CustomIQ cross‑platform measurements comparing December 2024 to December 2025. The company positions this as part of a broader “AI Intelligence” measurement effort that aims to track where people actually go to use assistants across devices.
For reashifts closely, the report confirms two clear dynamics: (1) mobile has become a primary access point for many assistant interactions, and (2) desktop remains essential for longer, more complex sessions, with marked concentration of desktop gains attributable to ChatGPT.

---

What Comscore reported — the key figures​

Here are the press‑release highlights presented in Comscore’s ranked snapshot for December 2025 (top items verified against Comscore’s release):

• Total mobile unique visitors (leading AI assistant destinations): 54.3M (↑107% YoY).
• Total desktop unique visitors (leading AI assistant destinations): 83.0M (↑18% YoY).

Top ranked destinations on mobile (Dec 2025):

• OpenAI ChatGPT — 34.5M (↑84% YoY).
• Google Gemini — 12.8M (↑137% YoY).
• Microsoft Copilot — 10.6M (↑246% YoY).
• Perplexity — 4.7M (↑265% YoY).
• Meta AI — 1.3M (↑50% vs. May 2025).

Top ranked destinations on desktop (Dec 2025):

• OpenAI ChatGPT — 56.4M (↑83% YoY).
• Microsoft Copilot — 33.4M (↓28% YoY).
• Google Gemini — 12.3M (↑648% YoY).
• Perplexity — 1.4M (↑516% YoY).
• Anthropic Claude — 1.1M (↑297% YoY).

Comscore frames the shift as a structural movement toward mobile for routine, on‑the‑go workflows while desktop remains the locus for extended, multi‑step tasks. That framing — and the specific numbers — are central to the interpretation in this piece.

---

Why the numbers matter: separating scale from growth rate​

The most important editorial point to understand is how to read percentage growth versus absolute scale.

• A +107% YoY increase on mobile is a fast growth rate, but it starts from a smaller base than desktop did historically for many assistants. Growth rates are therefore directional signals of momentum, not direct indicators of who currently controls the market.
• Absolute unique visitor counts (e.g., ChatGPT’s 56.4M desktop visitors) identify incumbents that still hold broad reach and habitual user bases. Comscore’s snapshot confirms ChatGPT’s continued dominance in absolute desktop reach while mobile share is becoming more distributed.

Independent trackers and app‑intelligence firms support the broad direction Comscore describes: mobile downloads and engagement for AI apps surged through 2025 (Sensor Tower reports major download and revenue spikes for generative‑AI apps), and referral telemetry shows ChatGPT continuing to dominate as a referrer to websites. Those external signals corroborate Comscore’s directional conclusions even though the underlying metrics—downloads, referrals, deduplicated unique visitors—are not identical and measure different behaviors.

---

Cross‑checks and corroboration: what other measurement firms say​

Comscore’s snapshot is useful, but measurement firms answer different questions. Here’s how Comscore’s claims line up with other reputable sources:

• Sensor Tower’s 2026 State of Mobile report highlights massive download and time‑spent growth for generative‑AI apps in 2025, noting ChatGPT’s extraordinary download and revenue gains (downloads up strongly, IAP revenue surging), which supports the finding that mobile usage for assistants is growing fast. That validates Comscore’s mobile‑first observation in a complementary metric (app installs and in‑app monetization).
• StatCounter’s referral tracking shows ChatGPT sending the majority of chatbot referrals to websites (roughly 80% in mid‑2025), which aligns with the notion that ChatGPT captures outsized desktop referral share even as mobile reach expands for others. This strengthens the claim that ChatGPT remains the major desktop/public‑web force for assistant‑originated traffic.
• TechCrunch and other industry outlets have previously reported app‑store milestones for ChatGPT (most downloaded app in certain months) and observed that distribution and integration (Google on Android/Chrome; Microsoft via Office/Windows) are important drivers for Gemini and Copilotduct dynamics picture consistent with Comscore’s ranking shifts.

Put simply: the cross‑checks don’t replicate Comscore’s unique‑visitor tabulation one‑for‑one, but they do corroborate the core story — mobile adoption is accelerating and ChatGPT retains dominant influence, especially in desktop/referral channels.

---

Methodologhis before you plan​

Comscore’s press release discloses that figures come from CustomIQ Dec 2024 & Dec 2025 and that the measurement is a person‑based deduplicated view spanning mobile and desktop. That methodology has advantages (cross‑device stitching, person‑level reach) but also important tradeoffs. The main points to keep in mind:

• Panel vs. telemetry: Comscore’s person‑panel approach can better capture cross‑device journeys and deduplicated unique users, while server‑side telemetry (app stores, backend logs) measures installs, sessions, or probes of endpoints. Each approach answers a different question; they are complementary, not interchangeable.
• Category definitions matter: How Comscore classifies “AI assistant destinations” (native apps vs. embedded assistive surfaces vs. search pages with AI overlays) will materially change totals. The public press release summarizes results but the full slide deck or dataset (gated by Comscore) contains the classification details. Treat top‑line numbers as an accurate summary of Comscore’s definition, not as a universal canonical truth across all measurement systems.
• Growth percentages on small bases can mislead: Triple‑ or quadruple‑digit growth rates for niche entrants sound dramatic but may arise from very small initial bases. Always compare the percentage to the absolute unique‑visitor count. Comscore’s release lists both, which helps tether the percentages to scale.
• Time windows matter: Comscore compares December 2024 to December 2025. Monthly snapshots can miss seasonal or product‑release spikes; consider longer rolling windows for strategic planning.

If you need to make procurement or engineering decisions based on these numbers, request the full Comscore presentation (they gate it) and specify the product‑and‑device classification you care about.

---

Platform‑level analysis: who’s gaining and why​

ChatGPT (OpenAI)​

• Strengths: dominant absolute reach on desktop and web referrals, strong mobile app performance and stickiness based on app‑intelligence reports. ChatGPT remains the primary discovery/referral gateway in many public‑web datasets.
• Risk/Weakness: As the market matures, retention and product personality tradeoffs (safety vs. spontaneity) may influence engagement patterns for certain use cases. Large incumbent scale also draws regulatory and scrutiny risk.

Google Gemini​

• Strengths: Rapid growth fueled by deep distribution (Chrome, Android, Workspace). Comscore shows large percentage increases on both mobile and desktop, consistent with Google’s OS/app surface area advantages.
• Risk/Weakness: Distribution can deliver reach quickly, but retention depends on product fit and usefulness. There’s also regulatory scrutiny about default placements and antitrust concerns.

Microsoft Copilot​

• Strengths: Enterprise provisioning and bundling with Microsoft 365/Windows create a powerful adoption channel — especially for productivity workflows that favor desktop. Comscore’s desktop numbers reflect that enterprise channel’s impact.
• Risk/Weakness: Enterprise pushes can produce installs that don’t always correspond to high consumer‑style engagement; organizations care about governance, security, and contractual assurances.

Perplexity, Anthropic Claude, Meta AI and others​

• Strengths: Niche positioning (research‑oriented retrieval, privacy emphasis, social integrations) yields high percentage growth from smaller bases, indicating opportunity in specialized verticals.
• Risk/Weakness: Smaller absolute footprints mean less influence over referral pipelines and lower likelihood of being the default choice for mass consumer discovery.

---

Publisher, advertiser, and platform implications — practical takeaways​

Comscore’s numbers are a diagnostic for three immediate business shifts:

• Publishers: Expect a sustained need to optimize for provenance and structured data. If assistants increasingly synthesize answers, publishers that supply verifiable, machine‑readable content (structured data, APIs, canonical knowledge blocks) will have higher odds of being cited by assistants. Puther being citable by assistants moves conversions.
• Advertisers / Media Buyers: Traditional last‑click attribution undercounts assistant influence. Brands must run experiments to measure whether appearances inside assistants (as citations or in‑app integran lift differently from paid search or display.
• Platforms & Product Teams: Design for cross‑device continuity and think in “assistant‑first” flows. If users start on mobile and finish on desktop (or vice‑versa), product metrics and funnels must be instrumented to capture those journeys.

Practical checklist for product and measurement teams:

• Inventory high‑value pages and add machine‑readable schemas and APIs.
• Design experiments to test “assistant preseels.
• Extend attribution windows and capture “assisted by AI” signals when feasible and privacy‑compliant.
• Build provenance metadata and sign responses where possible to resist manipulation.

---

Risks and governance: provenance, manipulation, privacy, and resilience​

The Comscore snapshot raises several near‑term and medium‑term risk areas every IT, security, and compliance team must consider:

• Provenance and manipulation: If assistants rely on a narrow set of domains as the basis for synthesized answers, that concentration can be gamed. Publishers and platforms should invest in verifiable metadata and signed feeds to reduce spoofing risk. Comscore signals attention to domain concentration, but the press release does not publish the domain‑rank tables publicly — this is gated in their full deck. Treat such domain‑mix claims as plausible but requiring verification.
• Privacy and training exposure: Many consumer tiers continue to use user inputs for model training. Enterprises and regulated verticals should insist on enterprise plans with non‑training clauses, data residency controls, and contractually defined retention.
• Operational dependence and outages: Heavy reliance on a single public assistant for business workflows creates continuity risk. Design fallback routes, multi‑vendor strategies, and cached knowledge resources.
• Factual risk (hallucinations): LLMs still hallucinate. For legal, financial, or operational decisions, require human verification and prefer citation‑first or RAG (retrieval‑augmented generation) setups that ground outputs in auditable sources.

Be explicit internally about what classes of data can never be sent to public assistant enterprises, prefer managed connectors, conditional access, SSO/MFA, and enterprise subscriptions that offer non‑training and compliance guarantees.

---

For WindowsForum readers (IT pros, sysadmins, power users) — an action plan​

If you run IT, security, or developer teams inside Windows shops, here’s a pragmatic, prioritized plan distilled from Comscore’s findings and independent corroboration:

• Inventory every point where assistants are used or can be used (browser extensions, plugins, macros, Power Automate flows). Classify by sensitivity.
• Enforce DLP and network controls to block prompt leaks of classified data to public tiers.
• Prefer enterprise / managed plans with contractual assurances (non‑training, data residency).
• Pilot RAG connectors that surface on‑prem knowledge to private LLMs or enterprise assistants for critical workflows.
• Implement logging, audit trails, and human‑in‑the‑loop verification for outputs used in customer communications or legal documents.
• Design graceful degradation strategies when public assistants are unavailable (local search, cached knowledge bases, fallbacks in PowerShell/Power Automate).

These steps follow a standard security lifecycle: discover → classify → pilot → harden → educate.

---

What Comscore doesn’t (yet) resolve — open questions to watch​

• Domain citation mix: Comscore signals concentration but the press release does not publish the domain‑level tables that would let publishers verify exact citation shares. Until the dataset or slide deck is public, treat domain‑mix claims as plausible but unverified.
• Session depth vs. reach: Comscore reports deduplicated reach (unique visitors) but does not publish session‑length distributions in the press release. Reach is necessary but insufficient to judge engagement intensity.
• Regional slices and verticals: The headline is global; many product decisions require country/vertical slices. Teams should request the full CustomIQ exports for targeted planning.
• Metric alignment across vendors: As always, reconcile definitions when combining Comscore (person‑panel reach) with app‑intelligence (downloads/MAUs) or referral telemetry (StatCounter).

---

Bottom line and strategic implications​

Comscore’s December 2025 snapshot is a useful, third‑party confirmation that assistants are moving beyond early adopter phases into mainstream use patterns — mobile is scaling fast and is now a primary access channel for many assistant interactions, while desktop remains the locus of deeper, extended sessions. Those trends are supported by app‑intelligence and referral data from other firms (Sensor Towh collectively paint a consistent industry narrative: distribution and embedding (preloads, enterprise provisioning) move reach quickly; incumbent scale (ChatGPT) continues to shape referral flows and public web influence.
If you run a digital property, an ad plan, or a Windows‑centric enterprise, treat Comscore’s findings as a measurement signal that should trigger three priorities:

• Make your content provable and citable (APIs, structured data).
• Revisit attribution and run A/B experiments that measure assistant presence.
• Harden governance and fallback operations for critical workflows that increasingly rely on external assistants.

---

Final verdict: opportunity and risk, side‑by‑side​

Comscore’s snapshot delivers an important pragmatic insight: assistants are now platform(s) — an additional discovery and productivity layer that sits alongside search, apps, and traditional referrals. That creates real opportunities for publishers, brands, and IT teams to be present in new UX surfaces — if they act quickly to supply verifiable, machine‑readable signals and govern usage carefully.
But the same forces that create opportunity also create systemic risk: provenance gaps, manipulatability if citation sets concentrate, privacy and training exposure, and operational dependence on a small set of services. Plan for both sides of that ledger: exploit the reach and integration benefits while building the technical and contractual controls that make assistant‑enabled workflows safe, auditable, and resilient.
Comscore’s press snapshot is an actionable wake‑up call for 2026: assistants aren’t a fad — they’re a new access channel. Your next step is to translate that measurement signal into product, measurement, and governance workstreams that protect customers and capture the upside.

---

Source: Comscore AI Assistants Head into 2026 on a High Note: Comscore Reports Triple-Digit Growth on Mobile