AI Advice Changes With Workplace Power
Artificial intelligence tools are increasingly being used as private workplace coaches, helping employees deal with difficult managers, colleagues and workplace conflicts. But a new report suggests that the quality of this advice may depend significantly on where an employee sits in the workplace hierarchy.
Research by Cloverleaf found that AI systems gave less constructive advice to employees with less workplace power, often siding with them instead of helping them understand the other person or repair the relationship.
Quality of Advice Fell 40%
The study found that when the person seeking advice had less authority, the average quality of AI coaching fell from 3.5 out of 5 to 2.1, a decline of roughly 40%.
The researchers said the difference was more strongly associated with power dynamics than with the personalities involved in the workplace conflict.
When the person asking for advice held authority over the other employee, AI models were more willing to challenge their interpretation of events and assign some responsibility to them.
But when the employee had less power, the models were more likely to accept that employee’s version of events and recommend protective measures.
Bosses Were Often Cast as the Problem
The pattern became particularly pronounced when conflicts involved managers.
Across two scenarios involving a boss, AI models described the boss as the problem in 60% of responses, while providing a more sympathetic interpretation of the boss in only 3% of responses.
The researchers described this as a roughly 20-to-1 tilt in how the models interpreted the conflict.
This raises concerns about AI reinforcing an employee’s initial assumptions instead of helping them examine the situation from multiple perspectives.
AI Rarely Encouraged Relationship Repair
Another striking finding involved the kind of solutions AI recommended.
Across 75 conversations, the models generated 638 distinct pieces of advice. Only three encouraged genuine investment in repairing the workplace relationship.
The remaining 635 suggestions focused largely on surviving, managing or winning the situation.
In conflicts involving managers, 27 of 30 responses raised leaving the job as a legitimate option. Suggestions included updating a résumé, exploring another team or considering whether the workplace was suitable.
Why This Matters
Employees may increasingly turn to AI precisely when they are frustrated and unwilling to immediately approach a manager or HR department.
That makes the nature of AI advice important. Advice that simply validates an employee’s interpretation could potentially escalate workplace conflicts rather than resolve them.
The researchers stressed that the findings are not an argument against using AI for workplace support. Instead, they suggested that AI coaching should be evaluated on whether it encourages self-awareness, accountability, understanding of others and relationship repair.
The Study Has Limitations
The researchers also cautioned that the findings are based on only 75 conversations and reflect specific AI model versions tested on a particular date. AI behaviour can change as models are updated.
Still, the research highlights an important challenge as AI becomes a workplace adviser: being supportive does not necessarily mean taking the user’s side.
Summary: A Cloverleaf study found that AI workplace coaching became roughly 40% less constructive for employees with less power, with models more likely to side with them and recommend self-protection. The research also found that AI rarely encouraged relationship repair, raising questions about whether workplace AI should be judged by how well it promotes accountability, empathy and constructive conflict resolution.
