When workplace conflicts become difficult, employees are increasingly turning to artificial intelligence for advice. But a new study suggests that leading AI models may be more likely to validate frustration and suggest leaving than help workers repair difficult relationships.
The finding comes from Cloverleaf Labs, the research arm of workplace relationship platform Cloverleaf, in its report AI as a Workplace Coach: What leading LLMs tell employees in conflict.
The study tested five leading large language models against five common workplace conflicts involving managers and colleagues. Researchers created 75 conversations, with each scenario run three times per model, and evaluated the responses across five dimensions of relational intelligence: self-awareness, accountability, other-awareness, specificity, and relational repair.
The results point to a significant gap between AI’s ability to provide actionable advice and its ability to help people navigate the interpersonal dynamics behind workplace conflict.
AI favors protection over repairing relationships
Across the 75 conversations, the models generated 638 distinct pieces of advice. Only three encouraged what the researchers considered genuine investment in repairing the relationship.
The report found that the models instead tended to focus on helping employees win, survive or manage the situation.
The pattern was particularly pronounced when the conflict involved a manager.
In the two scenarios involving a boss, 27 of 30 responses raised leaving the job as a legitimate option. Every model suggested leaving at least once.
Cloverleaf also found that models characterized the boss as the problem in 60% of responses, while offering a more generous interpretation of the manager’s behavior in only 3% of cases.
The overall scores reinforce the finding. Every dimension evaluated by the researchers fell below the neutral midpoint of 3 out of 5. Relational repair averaged 2.4, while specificity scored lowest at 2.2.
The findings are particularly relevant as workers increasingly use AI outside traditional productivity tasks.
The report cites research showing that 93% of workers have used AI to prepare for a conversation with their boss, while 49% said AI was more emotionally supportive than their manager.
That creates a new role for large language models: not just workplace assistants, but private advisers employees can turn to when they are frustrated with a colleague or manager.
The study also found that the quality of relational coaching changed when employees had less power than the person they were discussing.
According to Cloverleaf, relational coaching dropped by roughly 40% when the employee had less power in the relationship. In those situations, models were more likely to affirm the employee’s negative interpretation rather than challenge it or encourage consideration of the other person’s perspective.
Cloverleaf’s researchers designed the scenarios around two well-intentioned people with different working styles. Each person’s strengths could appear as weaknesses when viewed through the frustration of the other person.
The results suggest that current models often lean toward reinforcing that interpretation rather than helping employees reconsider it.
There was also considerable inconsistency between individual responses. One model produced the highest individual score in the study, 22.8 out of 25, while also producing one of the lowest scores, 7.4 out of 25.
For Cloverleaf, the issue is therefore not simply whether AI gives useful advice. It is whether that advice accounts for the relationship between the people involved.
“Even if it starts out introducing different ideas, the moment you indicate your preference, it tells you it’s a great idea. That’s not a thinking partner but a robotic affirmation,” said Kirsten Moorefield, Chief Strategy Officer at Cloverleaf.
The company argues that the answer is not to remove AI from workplace conversations. Instead, it says AI coaching should be held to a higher standard—one that considers self-awareness, accountability, awareness of the other person, specific guidance and the possibility of repairing the relationship.
As employees increasingly use AI to navigate sensitive workplace situations, the study suggests that the next challenge for workplace AI may be less about generating answers and more about knowing when not to simply agree with the person asking the question.
