Focusing on productivity gains or gameplay can be problematic, finds study
Since artificial intelligence (AI) has been introduced in the workplace, many organizations have faced ongoing questions about how to encourage employee adoption, especially when employees experience fear, uncertainty, or resistance
However, new research from university professors in the United States shows that the way AI is framed might play a key role in employees’ drive to adopt it. The study finds that framing matters not just in whether employees use AI, but in how creatively and collaboratively they engage with it.
“There’s a tendency to frame and position AI purely as a productivity enhancement tool, and that can backfire,” says Arvind Karunakaran, an assistant professor at Stanford University’s School of Engineering in California and one of the researchers behind the study.
“Framing it in a way that makes sense for the people using it — what they are gaining, not just monetarily, but in terms of skills and responsibilities — is essential. The core question to answer for the employee is: 'What’s in it or me?’”
The 2026 study was published in the Social Science Research Network and included professors from the University of Virginia and Dartmouth College.
Two law firms, two different approaches to AI adoption
To understand the employee view and association with AI, Karunakaran and his colleagues examined two workplace studies — and it is in the second study where the effect of framing becomes clear.
Over a two-year period, he and his colleagues observed two law firms with the same structure, similar pay, and the same AI tool — dubbed the “law bot.” Each workplace was allowed to decide how to use the tool on their own terms.
The first division had managers pitch the law bot to paralegals as a way to make work more efficient through drafting documents faster and quicker turnaround. The second division had managers ask paralegal employees to think about how to use the bot to free them up from routine tasks and give them time for more interesting work.
That led paralegals in the second division to not only experiment more freely with the tool, but have 58 per cent higher usage than those in the first division. They also began using it for legal research, case analysis, and sharing tips on how to use it.
The results of the second division surprised even Karunakaran.
“I was expecting some of the variation. What was less surprising was that the productivity messaging didn’t really work,” he says. “The second division, where we saw a lot more AI usage, used a different type of framing — not productivity. Without the advantage of retrospection, the signaling there also reads as, ‘You’re going to do more work’ or ‘You’re going to take on more.’ But it did work, and that was a bit of a surprise.”
The reason it worked, he says, comes down to a key distinction.
“It’s not that AI is being used to increase the workload,” he says. “It’s used to do more higher-order work that is actually aligned with what workers already wanted to do. That is the nuance.”
AI adoption stalls when job criteria does not keep up
The framing only works if the metrics follow, Karunakaran says.
When other divisions inside the same firm tried to replicate the same job enrichment of messaging, he says it worked initially — allowing people to do more advanced work — but that only lasted in the short term.
“As long as those divisions did not change the evaluation criteria for paralegals — which was mostly based on a caseload throughput metric — usage and adoption declined,” he says. “There’s often a lot of early adoption, followed by a drop-off or decline, partly because everything else hasn’t changed along with it.”
The AI gameplay trap that costs organizations
While experimentation can be a great way to help employees engage with AI, Karunakaran says there is something employers want to avoid relying on — AI gameplay.
This is a metric, such as token count, given when people use AI as a way to incentivize workers. The study finds that when workers are focused on hitting those numbers, it can lead to them not properly cross-checking work — leaving behind briefs that created errors for the rest of the team to clean up.
“They were trying to hit that mark. They were worried their numbers would drop because they were spending more time on complex work,” he says.
Gameplay, Karunakaran says, can be a great way to create buzz — but for the long term, it is not going to stick. The focus instead, he says, should be on allowing workers to experiment in a way that is recognized.
“They shouldn’t feel they’re doing all this experimentation on their own time — invisible work they’re not being rewarded for,” he says. “It needs to be woven into the job description or performance metrics and rewarded accordingly. If that doesn’t happen, many of these efforts fizzle out.”
What makes AI adoption last
Through the research results, Karunakaran outlines three things' employers need to do to make AI adoption hold.
The first is the messaging around AI and what employees can gain from the tool beyond productivity. If job enrichment is what employers want employees to take away from AI tools, he says they need to back it up — which leads to the second point.
Instead of focusing heavily on AI fluency or AI literacy — which many companies are doing by bringing in external vendors — Karunakaran says employers should move from an AI technology-first approach to a more task-first reskilling model.
For example, rather than only training paralegals on how to use AI tools better, employers should also train them on what it takes to do great legal research, then weave in how AI supports that goal — rather than starting with the technology.
The last is redesigning work, so AI fits the metrics. “You can’t keep using an older metric that reflects an older job," he adds.
“The question isn’t just how to do existing work faster — it’s what else becomes possible. What more complex, advanced work can individuals and business units now take on? That is the bigger mindset shift — from acceleration to expansion.”
