What Drives People to Actually Use ChatGPT at Work

The Hidden Reason People Actually Use ChatGPT at Work

It turns out that the most important factor in whether someone will use ChatGPT at work has almost nothing to do with how smart the AI is. It has to do with whether they trust the person who told them about it.

Mark Anthony Camilleri, a researcher at the University of Malta, surveyed 654 people who had used ChatGPT and found something counterintuitive. The strongest predictor of whether someone expected ChatGPT to be useful was not its speed, its accuracy, or its ability to write a passable email. It was something called “source trustworthiness” (Camilleri, 2024). People who learned about ChatGPT from a colleague they respected, a manager they believed in, or an expert they followed were far more likely to believe the tool would actually help them do their jobs.

This is not how most companies roll out new software. They send an email. They hold a training. They assume that if the tool is good, people will use it. But Camilleri’s research suggests that the social context around the technology matters more than the technology itself. The messenger is the message.

What Actually Predicts Whether Someone Will Use ChatGPT

Camilleri built his survey around two classic technology adoption models: the Technology Acceptance Model and the Unified Theory of Acceptance and Use of Technology. These frameworks have been used for decades to predict why people adopt everything from spreadsheets to social media. But Camilleri added a twist. He also measured “source trustworthiness” and “perceived interactivity” how much the user felt like they were having a real conversation with the AI.

The results, analyzed using a statistical technique called partial least squares structural equation modeling, showed three major drivers:

• ▸Source trustworthiness had the strongest effect on performance expectancy (how useful someone thought ChatGPT would be).
• ▸Perceived interactivity had the strongest direct effect on intentions to actually use the tool.
• ▸Performance expectancy itself was a significant predictor of use intentions, but it was partly shaped by trust in the source.

This means that people don’t evaluate ChatGPT in a vacuum. They evaluate it through the lens of who recommended it. If your boss says “try this,” you try it. If a random LinkedIn post says “try this,” you might not.

Why Trusting the Source Matters More Than Trusting the Tool

Here is the part that should make every product manager and IT director stop and think. Camilleri found that source trustworthiness influenced performance expectancy more than any other factor in the model. That includes the user’s own prior experience with technology, their perception of how easy ChatGPT was to use, and even their general attitude toward AI.

Think about what that means. A person who has never used a chatbot before, who is skeptical of AI, and who finds the interface confusing might still expect ChatGPT to be useful if they heard about it from someone they trust. Meanwhile, a person who is tech-savvy and curious might dismiss the tool entirely if the recommendation came from a source they do not respect.

This is not just about ChatGPT. It is about how humans process information in an age of information overload. We cannot evaluate every tool ourselves. We rely on social shortcuts. We ask: “Who told me about this? Do I trust them? Are they credible?” And that answer shapes everything that follows.

Camilleri’s survey captured this dynamic clearly. The participants who reported high source trustworthiness also reported significantly higher performance expectancy. They believed ChatGPT would help them write better, research faster, and solve problems more efficiently. And that belief, in turn, predicted whether they actually used the tool.

The Interactivity Paradox: Why Feeling Heard Matters More Than Being Right

The second major finding is subtler but just as important. Camilleri found that perceived interactivity how much the user felt like they were having a back and forth conversation with ChatGPT had a direct and significant effect on intentions to use the tool (Camilleri, 2024). This effect was actually stronger than the effect of performance expectancy.

This is strange. You would think that people use ChatGPT because it gives them good answers. But Camilleri’s data suggests that the experience of interacting with the tool the feeling of being heard, the sense that the AI is responding to your specific input matters just as much, if not more, than the quality of the output.

This aligns with something researchers have known for decades about human computer interaction. People prefer systems that feel responsive, even if they are not perfect. A chatbot that acknowledges your question, asks clarifying follow ups, and adapts its tone feels more useful than a chatbot that gives a perfect answer but feels robotic.

Camilleri’s finding adds a layer of practical urgency. If companies want employees to use ChatGPT, they should not just focus on improving the accuracy of the model. They should also focus on making the interaction feel more natural, more conversational, and more human. The tool that feels like a good listener will win, even if it is not the most accurate.

What the Research Does Not Prove

Before you start redesigning your company’s AI rollout strategy, a few caveats. Camilleri’s study is based on a survey of 654 people who had already used ChatGPT. That means the sample is self selected. These are people who were curious enough to try the tool in the first place. The findings might not apply to people who have never used ChatGPT and have no intention of doing so.

The study also relies on self reported data. People say they intend to use ChatGPT, but intention does not always translate into behavior. Camilleri measured “intentions to use,” not actual usage over time. It is possible that people overestimate how much they will use the tool, especially if they feel social pressure to appear tech savvy.

The research also does not tell us whether ChatGPT actually improves productivity. It tells us what makes people think it will. Those are different things. A person might trust their boss, try ChatGPT, and still find it useless. The study does not track outcomes.

Finally, the model explains about 60 percent of the variance in intentions to use ChatGPT. That is good for social science, but it means 40 percent of the reason people use the tool is still unknown. There could be other factors that matter more than trust or interactivity, such as personality traits, job requirements, or organizational culture.

How This Changes What Companies Should Do

If Camilleri’s findings hold up in future research, they suggest a very different approach to rolling out AI tools in the workplace. Most companies currently focus on the technology. They buy the best model, integrate it into their existing systems, and train employees on how to use it. But Camilleri’s data suggests that the social infrastructure around the technology matters more.

Here is what that looks like in practice:

• ▸Identify trusted sources within the organization. Do not send a company wide email about ChatGPT. Identify the managers, team leads, and informal influencers that employees already trust. Have them demonstrate the tool in small group settings. Let them answer questions. Let them model enthusiasm.

• ▸Make the first interaction feel like a conversation. Camilleri’s finding about interactivity suggests that the onboarding experience matters. If a new user types a question and gets a long, formal, impersonal response, they are less likely to come back. If the AI asks clarifying questions, acknowledges uncertainty, and adapts its tone, the user is more likely to engage.

• ▸Do not lead with accuracy claims. Companies love to say “our AI is 95 percent accurate.” But Camilleri’s data shows that accuracy is not the main driver of adoption. Trust and interactivity are. Lead with stories about how a trusted colleague used the tool to solve a real problem. Let people experience the tool in a low stakes setting before you ask them to rely on it for important work.

• ▸Measure trust, not just usage. Most companies track how many employees use a new tool. But Camilleri’s model suggests that the quality of the recommendation matters. If you see low adoption, do not immediately blame the tool. Ask: Who is recommending it? Are they trusted? Are they credible? If the answer is no, fix that before you fix the technology.

• ▸Accept that some people will never adopt. Camilleri’s model explains a lot, but not everything. Some people simply do not trust AI, no matter who recommends it. That is fine. Forcing adoption through mandates or performance reviews can backfire. The goal should be to create conditions where adoption happens naturally, not to force it.

What This Actually Means

• ▸The person who tells you about ChatGPT matters more than the ChatGPT itself. If you want people to use AI at work, invest in trusted messengers, not just in better models.

• ▸ChatGPT’s biggest competitor is not another AI. It is the human tendency to ignore recommendations from people we do not trust. Companies that understand this will win. Companies that do not will wonder why their expensive AI subscription is collecting digital dust.

• ▸Perceived interactivity is a feature, not a bug. If your AI tool feels like a vending machine that dispenses answers, people will not use it. If it feels like a conversation, they will. This is a design problem, not a technology problem.

• ▸Performance expectancy is built on social proof, not on benchmarks. People do not evaluate AI tools by reading white papers. They evaluate them by watching what their peers do. If the most respected person in the room uses ChatGPT, everyone else will want to try it.

• ▸The research is not a prescription. It is a starting point. Camilleri’s model is a snapshot of one moment in time, with one sample, using one methodology. The real test will be whether these findings replicate in other settings, with other tools, and with other populations. But for now, it is the best evidence we have that the human factor is the most important factor in AI adoption.

The next time you wonder why your team is not using ChatGPT, do not look at the AI. Look at the people around it. Look at who is talking about it. Look at who they trust. The answer is probably right there, sitting in a meeting room, wondering why nobody listens to their recommendations.