Google data: AI still complements work, not replacing jobs

Google researchers analyzed real-world usage patterns of Gemini—the company's large language model (AI system that understands and generates text)—by measuring how workers applied the tool to tasks documented in the O*NET database, a comprehensive inventory of U.S. work activities. The study classified interactions by setting a threshold: a work task counted as a 'Gemini task' only if it logged at least 25 related interaction attempts in the sample.

The results painted a picture of AI as an auxiliary tool rather than a wholesale replacement. Across the entire work-task database, only 21 percent of all work-related tasks reached this 'non-negligible' usage threshold. Breaking down by occupation, nearly 29 percent of jobs showed zero tasks with significant Gemini usage, and another 30 percent had fewer than one-quarter of their tasks receiving meaningful AI assistance. This left humans as the de facto owners of the vast majority of work components in most occupations. The study identified only 3 percent of occupations where Gemini was regularly consulted for at least three-quarters of relevant tasks—a cohort that included software quality assurance analysts and testers, human resources specialists, and document management specialists.

When the researchers examined the types of work Gemini users asked the model to handle, a clear pattern emerged: cognitive tasks (those requiring primarily thinking) accounted for 86 percent of all interactions by volume. Interpersonal tasks (like negotiation or coordination) and manual tasks remained underrepresented in Gemini usage relative to their actual prevalence in the workplace. The researchers concluded that 'AI is currently serving primarily as a complement to existing work' and that 'AI appears useful for a subset of tasks performed within occupations, but they do not currently appear to be comprehensively used for performing the work currently done by humans.' They acknowledged that future AI breakthroughs could alter this trajectory, but also flagged the possibility that workplace AI integration may stabilize into a model where workers and AI systems maintain 'a degree of complementarity.'

Google's ATLAS research team examined how deeply AI has penetrated the workplace by measuring Gemini interactions against the O*NET database of work tasks. The headline finding—that only 21 percent of work tasks saw significant AI usage—directly contradicts the automation hysteria that has dominated industry discourse. The researchers deliberately set a high bar: at least 25 related interactions in their massive sample. Even with this threshold, the vast majority of occupations fell well short.

The distribution reveals a stark split. Nearly three-tenths of all jobs saw zero meaningful Gemini adoption on any tracked task. Another 30 percent had less than one-quarter of their tasks receiving significant AI assistance. This means humans remained the primary decision-makers and executors for the overwhelming majority of work in most occupations. Only the smallest slice—3 percent of jobs—showed comprehensive AI integration across three-quarters or more of tasks. The researchers explicitly state that AI is 'currently serving primarily as a complement to existing work,' not as a replacement engine. While they acknowledge that future breakthroughs could shift this balance, they also note that workplace integration may settle into a stable hybrid model where workers and AI systems maintain complementary roles.