Is AI actually replacing workers? What Google's own usage data shows

For the past two years, the dominant narrative in the tech industry has been clear: AI is fundamentally transforming office work and taking over the tasks of millions of employees. That narrative has been repeated everywhere from conference keynotes to investor presentations by senior tech executives. But a new analysis based on Google's own usage data paints a far more cautious picture.

Researchers examined 15 million real interactions with Google's AI tools, analysing which work tasks those interactions corresponded to, how frequently they occurred, and to what extent the task was being automated. The goal was to build a picture grounded in actual usage patterns, rather than in survey data or executives' self-reported impressions.

The findings contradict what many expected: across the vast majority of tasks examined, AI is being used as a tool that supports an employee's work rather than one that takes over the task entirely. The analysis shows that most tasks in most jobs remain largely unaffected so far — a finding that directly contradicts the narrative of AI "sweeping away" jobs.

The impact is also unevenly distributed across types of work. Certain task categories, such as software development and content creation, show notably high usage intensity, while tasks requiring decision-making, customer relationship management, or physical operations see far more limited AI use.

Researchers explain this pattern through the distinction between "task-level automation" and "job-level automation." An AI tool might speed up or simplify a handful of an employee's weekly tasks without that employee's position disappearing entirely. Most jobs still involve elements — judgement, relationship management, contextual decision-making — that AI cannot easily take over.

The finding echoes a warning economists have long made: the impact of technological change tends to spread more slowly and unevenly than early forecasts suggest. Past waves of automation followed a similar pattern — specific tasks changed quickly, while full jobs remained largely stable for years, even decades.

Still, the analysis does not mean AI has had no effect. The researchers note that usage is growing rapidly, and that the impact in certain task categories could deepen over time. The findings represent a snapshot of the current moment; that picture could shift as AI tools' capabilities improve.

Another significant conclusion of the analysis is the gap it reveals between public statements from company executives and employees' actual day-to-day usage. Some senior executives have suggested that AI will radically shrink the workforce in short order, while the data shows such claims currently lack empirical grounding.

Experts argue that data of this kind could put labour-policy debates on a firmer footing. If automation's impact is being overstated, rushing into layoff waves or mass retraining programmes could produce policy decisions that do not match the actual need.

Researchers say repeating similar analyses over time will be one of the most reliable ways to track how — and how fast — AI's real impact on the workforce actually develops. For now, the data points to a gradual process of adjustment rather than a revolution.