The latest corporate snapshot of generative AI is both reassuring and revealing: most companies report measurable gains, but the market for the tools they use is sharply uneven — with OpenAI’s ChatGPT dominating usage, Microsoft’s Copilot punching above its weight thanks to office integration, Google’s Gemini trailing in enterprise reach, and Anthropic’s Claude underperforming relative to expectations.
In late 2025 a widely circulated Wharton Human‑AI Research survey of business leaders confirmed a major shift in how companies treat generative AI: what started as experimentation has become a line-item investment measured by ROI, governance, and integration. The Wharton findings show that a significant share of firms now track returns from generative AI projects and that nearly three out of four respondents are already reporting positive ROI from those investments. That positive‑ROI headline has shaped recent coverage and framed the debate about which AI tools businesses actually adopt. This piece synthesizes the Wharton snapshot and Business Insider’s reporting on the same data, verifies the most consequential claims against independent market trackers and industry reporting, and examines why some assistants are winning adoption while others lag. It also flags claims that cannot be independently verified and gives pragmatic guidance for IT teams and Windows‑centric organizations evaluating copilots, chatbots, and enterprise LLM deployments.
Source: Business Insider The AI tools used most by companies. There's a surprising winner and a shocking laggard.
Background
In late 2025 a widely circulated Wharton Human‑AI Research survey of business leaders confirmed a major shift in how companies treat generative AI: what started as experimentation has become a line-item investment measured by ROI, governance, and integration. The Wharton findings show that a significant share of firms now track returns from generative AI projects and that nearly three out of four respondents are already reporting positive ROI from those investments. That positive‑ROI headline has shaped recent coverage and framed the debate about which AI tools businesses actually adopt. This piece synthesizes the Wharton snapshot and Business Insider’s reporting on the same data, verifies the most consequential claims against independent market trackers and industry reporting, and examines why some assistants are winning adoption while others lag. It also flags claims that cannot be independently verified and gives pragmatic guidance for IT teams and Windows‑centric organizations evaluating copilots, chatbots, and enterprise LLM deployments.Executive summary of the Business Insider / Wharton coverage
- Wharton’s Human‑AI Research work shows rapid maturation: AI is moving from pilots to measurable programs, with many firms tracking ROI and adjusting budgets accordingly. About 46% of business leaders report daily use of generative AI, and roughly 72–75% of organizations report positive ROI on AI projects.
- Business Insider highlights the tools companies say they use most: ChatGPT tops the list; Microsoft Copilot ranks ahead of Google Gemini in enterprise adoption; Anthropic’s Claude shows surprisingly low adoption in these enterprise surveys.
- Independent traffic and market‑share trackers corroborate that ChatGPT leads by a wide margin in public usage and referral traffic, while Microsoft Copilot has shown strong growth month‑over‑month — frequently attributed to deep Microsoft 365 and Windows integration. Several trackers place Gemini and Claude far behind ChatGPT on global referral-share metrics.
Why the Wharton numbers matter (and what they actually say)
Measured adoption, not hype
Wharton’s reporting—and the broad media coverage that followed—marks a pivot from anecdotes to measurement. Where AI budgets and vendor pitch decks once dominated the conversation, companies are now focusing on:- Structured KPIs (productivity gains, time saved, error reduction),
- Governance and non‑training or data‑use guarantees for enterprise contracts, and
- Operational integration (how a model connects to tenant data, identity, and compliance tooling).
What the ROI claim does not prove
The Wharton headline should be read carefully. Survey‑based ROI claims reflect respondent experience and expectations; they are real and important, but they are not the same as audited financials showing net profit increases attributable solely to AI. Where Wharton’s data is strongest is in showing directional business value and adoption patterns, not in proving universal profitability across industry segments. Independent industry trackers and vendor filings show uneven results by sector, scale, and use case. Treat the Wharton ROI number as a trustable indicator of broad success, not a universal guarantee.The AI tool leaderboard: who’s winning and why
1) ChatGPT (OpenAI) — the dominant generalist
Market telemetry from multiple trackers places ChatGPT at the top of public usage and referral traffic. StatCounter‑derived snapshots and industry reporting in 2025 repeatedly show ChatGPT commanding a large majority of chatbot referral share — often in the 70–80% range depending on the dataset and measurement method. Comscore and other audience measurement firms also place OpenAI at or near the top for unique visitors and engagement. That dominance translates into real enterprise traction because scale brings ecosystem advantages: plugins, APIs, third‑party integrations, and a huge pool of users and developers who have already incorporated ChatGPT into workflows. Why enterprises use ChatGPT:- Broad capability set (writing, coding, summarization, multimodal features),
- Mature developer ecosystem and APIs for automation,
- Familiar product tiers that range from free to enterprise with admin tools.
2) Microsoft Copilot — the productivity ace
Microsoft’s Copilot shows striking enterprise adoption relative to its overall referral share because it’s embedded in productivity software organizations already pay for. Copilot’s advantage is distribution and contextual access: it can act inside Word, Excel, Outlook, Teams, and Windows with tenant controls managed by administrators. That integration reduces friction and increases day‑to‑day usage. Comscore’s mobile and audience analyses also show big growth rates for Copilot, reflecting the power of embedding AI inside the operating environment. Why Copilot is surge‑ready:- Deep Microsoft 365 / Windows integration,
- Enterprise governance through Microsoft Graph, Purview, and tenant assurance,
- Familiar licensing channels for IT procurement.
3) Google Gemini — multimodal strengths, slower enterprise conversion
Google’s Gemini brings powerful multimodal features and a one‑million‑token context capability in some model families, which is a technical differentiator for long‑document analysis. However, distribution into enterprise workflows has been harder to translate into daily usage compared with Microsoft’s Copilot. Part of Gemini’s challenge is that full value often requires deep adoption of Google Workspace; cross‑platform organizations or Windows‑first enterprises find Microsoft’s bundling easier to operationalize.4) Anthropic Claude — the surprising laggard
Anthropic’s Claude has been widely praised for safety design and long‑form reasoning capabilities, and industry observers expected faster enterprise adoption. Survey snapshots in Business Insider and other outlets show Claude’s usage lower than many insiders anticipated. Independent telemetry places Claude well below ChatGPT in referral share and active traffic, although Claude has pockets of traction in privacy‑sensitive use cases and long‑context workflows. Anthropic’s enterprise contracts emphasize non‑training guarantees, but that hasn’t yet produced ChatGPT‑scale usage in public telemetry. Important nuance: Claude’s enterprise traction is meaningful in contract negotiations and downstream embedded offerings, but public traffic-based measures understate behind‑the‑firewall deployments that enterprise customers keep private. Treat low public referral share as a real signal of comparative public usage, not definitive proof of enterprise irrelevance.Why the gap between expectation and adoption exists
Integration and distribution beat raw model quality
Two themes keep recurring in corporate procurement decisions:- Distribution & friction reduction — if a model is embedded into apps employees already use every day, adoption explodes. Microsoft's Copilot benefits from this. Google’s Gemini has the same potential inside Workspace but has had more trouble converting distribution into broad enterprise habit outside Google-centric shops.
- Governance & contractual clarity — firms buying AI for regulated work care as much about non‑training clauses, data retention, and auditability as they do about creative prowess. Anthropic and Microsoft both emphasize enterprise guarantees, but procurement dynamics (existing vendor relationships, licensing channels, security artifacts) can favor incumbents like Microsoft and OpenAI.
Trust, habit, and switching costs
Enterprises build processes around tools. That builds habit — teams learn prompts, templates, and verification steps. Switching to a differently integrated assistant is costly in time, training, and governance adjustments. That explains part of ChatGPT’s continued sticky dominance even as rivals advance technically.Risks, caveats, and unverifiable claims
What’s verifiable
- Survey findings (Wharton): high daily and weekly usage among managers, and a majority reporting positive ROI. Multiple outlets reproduced these takeaways from Wharton’s work.
- Public traffic trackers (StatCounter, Comscore, industry press): ChatGPT leads in referral traffic and unique visitor metrics in most datasets; Copilot shows rapid growth but is still behind ChatGPT on raw web referrals.
What remains uncertain or vendor‑asserted
- Vendor claims about absolute user counts, parameter sizes, training costs, or “billions” of unique users are often not independently auditable from public telemetry. These numbers are frequently repeated in marketing and press releases but should be treated with skepticism until backed by audited data. Examples include some press‑quoted user totals and dramatic growth percentages for regionally focused entrants. Those claims need independent verification. Flagged: treat vendor‑level totals and dramatic growth percentages as claims, not facts.
Operational risks companies face
- Hallucinations and factual drift: models still make confident but incorrect statements; enterprises must enforce human‑in‑the‑loop checks.
- Data leakage and training usage: before deploying consumer tiers, firms must verify enterprise contracts that guarantee non‑training or specific retention policies.
- Concentration risk: heavy dependence on a single provider creates fragility — outages or policy shifts can halt workflows. Multi‑vendor fallbacks are prudent.
Practical guidance for IT leaders and Windows-focused admins
When evaluating or scaling generative AI in an enterprise, treat the project as a product rollout, not a feature flip. The following steps offer a practical playbook.- Define measurable KPIs before rolling out pilots:
- Productivity (time saved per task), quality (error reduction), throughput (cases handled), and financial uplift (incremental profit or cost avoidance).
- Start small, measure, and iterate:
- Pilot in a contained function, instrument usage, and prove value against the pre‑defined KPIs.
- Enforce data governance:
- Require enterprise contracts with clear non‑training clauses for sensitive data and map data flows for compliance teams.
- Map integration points:
- If your organization is Microsoft‑centric, evaluate Copilot first for frictionless value; if Google Workspace is dominant, Gemini may be the natural choice.
- Plan for multi‑vendor resilience:
- Avoid single‑vendor lock‑in for mission‑critical workflows; maintain failover tools and scripts to reduce outage impact.
- Invest in training and role redesign:
- The human side — skills, morale, role clarity — predicts success. Build training and change management into the rollout budget.
- Monitor costs and quotas:
- AI tiers are frequently metered; set alerts and governance to avoid runaway bills.
What enterprise buyers should ask vendors (short checklist)
- Do you provide a contractual non‑training guarantee for enterprise data?
- What audit reports (SOC 2, ISO) and data residency options do you provide?
- How are admin controls, tenant grounding, and access policies implemented?
- What are your rate limits, expected latency, and enterprise pricing tiers?
- Can you provide references for customers in our industry with similar scale?
The competitive outlook: how this changes the vendor playbook
- Distribution as a moat: embedding assistants into productivity suites (Microsoft) or device ecosystems (Google on Android/Pixel) is a faster route to habitual enterprise usage than marginal model improvements alone.
- Specialization wins niches: companies that need citation‑forward, research‑grade answers may favor Perplexity‑style tools; safety‑focused, long‑form use cases may prefer Claude‑style offerings with long context windows and enterprise contracts.
- Open vs. hosted models: open‑weight models and private deployments will matter more where data sovereignty and IP control are essential; that dynamic benefits specialized cloud vendors and some open‑model projects.
Conclusion
The Business Insider story built on Wharton’s Human‑AI Research work offers a realistic, mildly optimistic account of enterprise generative AI in 2025: most companies that track ROI are seeing benefits, but the market’s tool adoption is concentrated. ChatGPT’s public and enterprise reach is dominant; Microsoft Copilot’s deep app integration gives it a productivity edge inside Office/Windows ecosystems; Google’s Gemini brings impressive multimodal technical capability but faces conversion challenges in mixed environments; and Anthropic’s Claude, while technically notable, lags in public adoption metrics. These conclusions are visible in survey reporting and corroborated by independent audience and traffic trackers. For IT leaders, the implication is clear: pick tools by use case and governance posture, instrument outcomes with rigor, and treat AI rollouts as product programs that require measurement, training, and contingency planning. Generative AI is no longer a hypothetical advantage — it’s operational capability. The winners will be teams that measure the outcome, control the risk, and integrate AI into the daily rhythm of work without surrendering accountability.Source: Business Insider The AI tools used most by companies. There's a surprising winner and a shocking laggard.
