AI-assisted performance reviews can’t replace human insight, report warns
A new report from Highwire, a coaching program run by Knopman Marks, finds that 78% of U.S. people managers have used AI tools for employee feedback or performance reviews in the past year — yet only 16% of employees were informed that AI played a role in their most recent review. The transparency gap is creating a psychological safety crisis: roughly two-thirds of employees report changing their workplace behavior to protect themselves, including avoiding risks, withholding ideas, and hiding mistakes.
Key facts
- 78% of U.S. people managers surveyed used AI tools for employee feedback or performance reviews in the past 12 months (Highwire/Knopman Marks, Aug 2026, n=300+ managers).
- Only 16% of individual contributors (n=700+) were told AI played a role in their most recent performance review.
- 54% of employees found AI-generated feedback more specific and actionable; 34% found it more generic; 32% found it less useful.
- ~Two-thirds of employees reported changing their behavior to self-protect, including avoiding reasonable risks (27%), delaying asking for help (27%), avoiding challenging responsibilities (23%), holding back ideas (21%), and hiding mistakes (16%).
- A July 2026 PYX Labs/Perceptyx study reportedly found AI models struggle with incomplete, emotional, or context-dependent signals.
- Radical Candor research found 73% of 600 U.S. employees had reported inaccuracies in AI-assisted work, but ~half said managers rarely or only sometimes acted on those reports.
- American Management Association findings indicate two-thirds of professionals say they are not getting frequent support from managers.
AI tools have quietly become a fixture of the performance review cycle at U.S. companies, but a new survey suggests the rollout has outpaced the transparency employees need to trust the process — with measurable consequences for development, risk-taking, and organizational learning.
Highwire, a performance conditioning coaching program operated by exam-training firm Knopman Marks, surveyed more than 300 people managers and over 700 individual contributors across the United States in August 2026. The findings, published October 8 in HR Dive, paint a picture of an AI adoption wave that has moved faster than governance norms can keep up with.
**Adoption is widespread; disclosure is not** Seventy-eight percent of managers said they have used AI tools to handle employee feedback or a performance review at least once in the past twelve months — a striking majority for a task long considered one of the most human-dependent in people management. Yet only 16% of individual contributors reported being told that AI played any role in shaping their most recent review.
Employee sentiment on AI-generated feedback was divided. Roughly 54% said they found it more specific and actionable than traditional feedback, while 34% called it more generic and 32% said it was less useful — suggesting the quality of AI-assisted reviews varies considerably depending on how managers engage with the output.
**A psychological safety crisis in the making** The opacity around AI use is triggering defensive behaviors that run counter to what high-performance cultures require. Although most survey respondents believe that making mistakes is critical to learning new skills, more than half said they feel they cannot make a mistake without it being held against them. As a result, approximately two-thirds reported changing their behavior to self-protect: 27% said they avoid reasonable risks, another 27% delay asking for help, 23% sidestep challenging responsibilities, 21% hold back ideas, and 16% admit to hiding or downplaying mistakes.
"When employees suspect a tool shaped their review, and no one has told them how, playing it safe starts to feel like the smartest move," Highwire co-founder and CEO Liza Streiff told HR Dive.
**The irreplaceable value of context** Streiff argued that the core limitation of AI in performance management is not sophistication but knowledge: AI models draw on aggregated training data, not lived experience with a specific employee. "Its knowledge is an amalgamation of millions of other people, essentially the textbook version of what feedback should sound like," she said. "It doesn't know how someone has grown over the year, and that context is the whole point of a review."
The distinction matters in high-stakes moments. A good manager, Streiff noted, can differentiate between a smart risk that did not pay off and a genuine careless mistake — a judgment call that requires relational context AI systems currently cannot replicate.
The critique is supported by outside research. A July 2026 study from PYX Labs, the research arm of employee-listening platform Perceptyx, reportedly found that while AI models handle themes with clear answers well, they struggle significantly with incomplete, emotional, or context-dependent signals — precisely the signals most relevant in a performance conversation.
**A disclosure and training imperative for HR** The solution Streiff outlined is less about limiting AI use and more about restoring transparency and redirecting the time AI saves. "HR can really help by making sure managers are open about how AI was used, and that the time it saves goes back into the conversation itself," she said.
Separate research adds urgency. A Radical Candor survey of 600 U.S. employees found that 73% had reported inaccuracies in AI-assisted work, yet roughly half said managers rarely or only sometimes acted on those reports. And findings from the American Management Association suggest two-thirds of professionals do not feel they receive frequent support from their managers — a baseline deficit that AI-assisted reviews could worsen if they further reduce meaningful human interaction.
**The bottom line for HR leaders** The Highwire report frames AI's role in performance management not as a binary question of use or non-use, but as a governance and communication challenge. Disclosure, manager training, and a deliberate commitment to human conversation are the variables HR teams can actually control. "Developing people will always require people," Streiff concluded. "We have to make sure we're not replacing the human development employees actually need with the version that's just easier to reach for."
Performance reviews sit at the heart of employee development, retention, and trust. As AI tooling quietly reshapes how feedback is generated, the disclosure gap exposed by this research creates a compounding risk: employees who do not understand how their review was produced are less likely to trust it, more likely to play it safe, and less likely to grow. For HR technology buyers and practitioners, the story reframes the AI-in-HR debate from 'does it work?' to 'how do we govern it?' — a shift with direct implications for vendor selection, manager training, and employee experience strategy.