Organizations are pouring unprecedented resources into artificial intelligence (AI). Billions of dollars are being spent on enterprise licenses, training programs, and new AI-powered tools designed to transform how work gets done. Yet a curious pattern continues to emerge. While some teams rapidly integrate AI into decision-making, problem-solving, and daily workflows, others barely move beyond experimentation.
The conventional explanation is straightforward: adoption begins with leadership. When leaders communicate a compelling vision, model desired behaviors, and encourage employees to embrace new technologies, adoption should follow. The influence of leadership is difficult to ignore. Gallup reports that employees are 2.1 times more likely to use AI frequently when their managers actively support its adoption, and 8.8 times more likely to believe AI helps them do their best work. When leaders embrace AI, employees are more likely to see its value for themselves.
Yet the reality is far less transformative than the headlines suggest. Gallup also reports that four in ten U.S. employees never use AI at all, and only three in ten have incorporated it into their work on at least a weekly basis. Revelio Labs, on the other hand, has found significant adoption gaps within the same organizations, with younger employees and knowledge workers embracing AI far more quickly than older workers and those in support roles.
Behaviours of trusted collegues
Researchers from Microsoft, in the HBR article Peer Influence Can Make or Break Your AI Rollout point to a different explanation. While leadership support creates the conditions for adoption, employees are often influenced more strongly by the behaviors of trusted colleagues. Individuals who observe coworkers experimenting with AI, applying it to real business challenges, and openly sharing their experiences are substantially more likely to become active users themselves.
Peer-to-peer learning accomplishes what formal training often cannot. It provides visible proof that AI is practical, safe, and valuable in real work. When employees see trusted colleagues using AI to solve problems and improve performance, adoption is far more likely. Peer influence was especially powerful when employees had a close, trusted colleague to learn from. Among employees in the top quartile of AI use, 88 percent described their local peers as highly influential. While only 50 percent of those in the bottom quartile said the same thing.
Our own research reveals a similar pattern. AI adoption is rarely distributed evenly across an organization. In the 1,100-person division of a large services company shown below, weekly use of formal AI tools clusters heavily in specific parts of the network, particularly in the center and upper-left region. The darker the node, the greater the frequency of use. Meanwhile, many employees on the outer edges of the network remain light or infrequent users.
Peers have a significant impact on AI adoption rates
Our findings are also consistent with Gallup's research. Employees are roughly twice as likely to adopt AI when their leaders use it, but they are nearly three times as likely when a trusted colleague does. That dynamic is clearly visible in the network below. Along the outer edges are multiple clusters of low use where AI adoption remains limited. In many cases, these individuals are surrounded by peers who are also infrequent users, reinforcing existing patterns of behavior.
Most organizational change efforts assume that information drives behavior. Leaders communicate priorities, provide resources, and explain why change is important. Yet behavior rarely changes simply because people receive new information. Behavior changes when the value is elevated or uncertainty is reduced.
A leader's message explaining the strategic importance of AI may be informative and even inspiring. However, it often remains abstract. The employee may still wonder how the technology applies to their specific work, whether they are using it correctly, or whether experimentation carries risks. By contrast, observing a trusted colleague use AI to solve a practical problem immediately changes the equation. The technology becomes tangible. The benefits become visible. What previously felt uncertain begins to feel achievable.
Proving the safety of the new ways of working
The difference is not simply access to information. The difference is social proof. People want evidence that a new way of working is effective, valuable, and safe. They look to those around them for cues.
This helps explain why AI adoption often varies dramatically across teams even when access to technology, training, and leadership support remains constant. The difference frequently lies within the social system rather than leadership support. In some pockets, knowledge spreads rapidly through relationships. Confidence grows. Adoption accelerates. Other pockets experience the opposite dynamic. Individuals experiment privately, discoveries remain isolated, and valuable learning never diffuses across the group.
From a network perspective, this distinction is critical. Formal leaders shape the environment in which adoption occurs, but informal networks determine how behaviors spread. Research on social influence has consistently shown that people are more likely to adopt new behaviors when they observe those behaviors being practiced by trusted members of their network. Influence travels through relationships, and trust amplifies that influence.
So, what can we do to elevate adoption:
- Be a visible experimenter: Let colleagues see how you use AI in your daily work. Adoption increases when people observe trusted peers using it successfully.
- Show practical value: Demonstrate how AI solves real business problems, saves time, or improves outcomes rather than talking about the technology itself.
- Share what you learn: Exchange prompts, use cases, and lessons learned with others. Peer-to-peer learning often spreads adoption faster than formal training.
- Reduce uncertainty: Create psychological safety for colleagues by openly discussing both successes and failures, helping others feel confident experimenting with AI themselves
In short: Model the behavior, demonstrate the value, share the learning, and reduce the fear.