78% of managers used AI for performance reviews — only 16% told employees

Seventy-eight percent of managers used AI to help draft, edit, summarize, or inform a performance review in the last year, according to professional development firm Highwire's survey of 1,034 corporate employees.

The figure lands at a moment when workplace AI adoption is no longer experimental. Nearly 70% of workers now use AI regularly, with little time allotted for formal upskilling, per a separate ZDNET-cited workforce study. Performance reviews — long an exercise in deliberative managerial writing — have become an AI-shaped process. Most of the people being reviewed have no idea.

What the survey measured

Highwire's research tracks how AI has entered the feedback loop between managers and their direct reports. The 78% figure is the headline. Beneath it sits a stack of results on how employees experience those AI-touched reviews:

• 54% of employees said feedback "has become more specific and actionable"
• 34% found feedback more generic
• 32% said it was less useful than before

A third of workers potentially pulling less value from reviews than they did pre-AI is not a marginal finding. The majority may be benefiting. A meaningful minority is not.

Highwire told ZDNET it did not ask managers what else they noticed about using AI. The firm declined to specify whether the omission was methodological, a length constraint, or an editorial choice. The result, regardless, is a partial map of effects — a survey that documents use but not consequences.

Why employees aren't being told

Only 16% of non-managerial employees surveyed said they were told AI influenced their review. That 62-percentage-point gap between manager practice and employee disclosure defines a structural transparency problem.

Highwire didn't isolate cause. A June Atlassian study cited by ZDNET points to one driver: owning AI use at work can backfire on the employee who admits it. If disclosing AI-assisted work carries social risk, fewer managers will volunteer that their review was machine-aided. The asymmetry persists regardless of whether disclosure is moral, regulatory, or performative.

How AI is being used — and how much the difference matters

The 78% headline doesn't distinguish between light and heavy AI use. That distinction carries weight.

An AI tool that formats or polishes a human-written observation is a different intervention from handing AI the underlying performance data and asking for a finished review. Highwire's instrument didn't disaggregate. That omission leaves the survey unable to answer the question most relevant to employees: how much of the feedback was actually generated by the manager?

The level of intervention plausibly shapes the outcome. A lightly formatted review still encodes managerial judgment. A fully AI-generated draft may flatten the context, the specificity, and the confrontation an employee needs to grow.

Rehearsing difficult conversations with AI

The survey's second major finding: nearly 1 in 4 employees now rehearse difficult workplace conversations with an AI tool.

Highwire frames this as a vacuum-filling effect. Only 44% of employees reported having their manager's support to practice a high-stakes conversation or other difficult workplace scenario. That is far below the 85% who say practice is key to getting it right.

The gap — 41 percentage points between those who value practice and those who get managerial support for it — is where AI has stepped in. Employees nervous about confrontation now turn to chatbots for rehearsal they cannot get from a human manager.

Where Synthesia's Sessions fits

Synthesia, the AI video company, has launched a feature called Sessions that places an AI avatar in the role of conversation partner. The product includes an "AI coach" that assesses the employee's performance afterward.

In product terms, Sessions is an automated, scaled version of the one-on-one training managers typically provide. Synthesia built the feature in response to enterprise customers who wanted broader access to those managerial touchpoints without expanding human bandwidth.

The company maintains Sessions is not a replacement for mentorship. Independent data on how effective the training experiences are for employees has not been published.

Both Highwire's research and Synthesia's product investment point in the same direction: AI is taking on a larger share of the simulation and rehearsal work that once happened between a manager and a report, even as the firms frame the trend through opposite commercial lenses.

Whose interests the data serves

Highwire's commercial position shapes how the survey results were framed. The company's business depends on companies needing human-led professional training — specifically the people skills AI cannot easily replicate.

The company's site states: "Highwire conditions the human skills that can't be automated."

That framing does not undercut the data. The 78% figure, the 16% disclosure rate, and the rehearsal gap are all concrete measurements. The conclusion the company draws from them tilts toward human-led training as the remedy. Synthesia, operating from a different commercial angle, sees the same gap and builds an AI tool to fill it.

Both companies are correct about the gap. They diverge on the cure.

What this means for AI at work

Two trends sit inside the survey and the Sessions release. AI is now a routine, usually invisible, layer in the manager-to-employee feedback chain. AI is also filling the practice gap managers have left open.

Some employees argue AI-generated reviews already qualify as "workslop" — a working term for AI-produced output that shifts cognitive effort onto the recipient. A less specific review is a tax on the employee's time. A rehearsal with a chatbot is one fewer conversation with a real manager.

The transparency problem is the most actionable piece. The 16% disclosure rate is a number an HR department, a chief people officer, or a regulator can move. Mandating disclosure would do little to slow AI adoption in feedback. It would change who knows.

The next wave of workplace AI tooling — Synthesia's and others' — is likely to land on the rehearsal gap before the disclosure gap. HR teams will not move quickly enough on transparency rules. Vendors are already moving on simulation.