On August 26, talent-assessment firm The Predictive Index released a July survey of 399 managers and 208 CEOs and business leaders across US industries, job levels and generations. The subject was the hard conversation: the performance review, the criticism, the exit.
Two findings belong next to each other. 72% of managers said public AI tools are useful for preparing those conversations. 44% said they had already entered employee names and performance details into them.
74% against 20%
Start with the gap in what each side believes.
74% of CEOs and business leaders said their managers are “very confident” handling tough people conversations on their own.
Managers answered differently. Only 20% said they prefer to prepare for a hard conversation alone, without a framework, a coaching guide, or input from HR.
Three quarters of managers need no help, by the leadership count. Four fifths of managers want help, by their own. That is not a confidence problem, it is a visibility problem. Anthony Belluccia, an I/O psychologist at PI, named it plainly: trusting a manager and knowing they are set up for a specific hard conversation are two different things, and managers are more aware of where they need support than their leaders are.
The survey also split out where the gap sits. 42% of managers said they know what they want to say and struggle with how to say it. The top three blind spots: delivering criticism constructively (22.8%), anticipating the employee’s reaction (19.5%), staying objective (14.3%).
None of those is knowledge. All three are judgment under pressure, and judgment under pressure is exactly what two decades of shrinking management-training budgets stopped funding. A WTW report last year pointed the same direction: only 1 in 5 organizations said their managers effectively provide feedback.
44% and 45%
Put an unmet need next to a free tool in the browser and the rest follows.
The striking part is a different pair of numbers. 44% of managers have typed employee names and performance details into public AI tools. Among the organizations surveyed, 45% have a formal written policy governing AI use in performance management.
Nearly the same size, measuring opposite things. One is how far the behavior has spread. The other is how far the rules have. Rules cover barely half; the behavior covers nearly half, and that is only the share willing to admit it on a survey.
PI also recorded that employers hold clear concern about employee information landing on public platforms. The space between that concern and a written rule is the current operating state: companies know the risk and have not written down that the answer is no.
The wording matters too. PI asked about public AI tools, meaning consumer chat products, not the licensed enterprise tenant with a data-processing agreement attached. The distinction is where the record lives. In one case it sits inside the company’s systems. In the other it does not.
The layer being outsourced is the one that resists codification
Earlier this month we covered the Stanford work that widened the young-worker employment gap to 19% and named the mechanism: knowledge that can be written into documentation, courses and standard procedure goes first, and tacit knowledge built on the floor holds longer.
The hard conversation is the textbook case of tacit knowledge. How to keep going when the other person’s face changes. How to stay level after being contradicted. There is no manual. There are only the several dozen times you have done it.
What this survey captures is the moment that layer starts being outsourced. Managers are not handing over the decision. Who gets cut and who gets what rating is still set by a person. What is being handed over is the phrasing, the pacing, the rehearsal, which is the part of those twenty minutes that actually took experience to acquire.
The sequence is worth noting. A role usually has its judgment tooled before it has its headcount cut. Middle management has been named repeatedly in the past two years of layoffs, generally under the heading of too many layers. A layer can only be compressed if what it uniquely provided can come from somewhere else. When three quarters of managers report that a general-purpose chat tool is useful for the hardest part of the job, that premise is being tested from the inside.
What it means for the person on the other side of the table
For the employee, this changes one concrete thing: the wording of the conversation may not have been formed inside any company system.
If the performance language, the improvement plan, or the exit script was drafted in a personal chat account, then the record an employee can request through company process and the actual origin of those words are in two different places. The file holds the conclusion. The reasoning sits elsewhere, in a place with no retention policy, no access route, and no exposure to a labor dispute process.
This is not a hypothetical edge case. Last week we covered Andon Labs’ AI store manager, which fired its first human after losing its own rulebook. There the AI made the decision and mislaid its basis. Here the direction reverses: the human decides, the AI drafts, and the basis never enters the company’s systems at all. Both land on the same result, which is that decisions are getting harder to trace.
The same month, the CNBC survey found that the most anxious workers are the occasional users, not the abstainers. The PI data fills in the other half. A meaningful share of those occasional users are managers, and they are reaching for the tool to handle the most uncomfortable part of their job.
Watch the 45% over the next year. States have already started requiring employers to declare whether a layoff is AI-related, and performance management is the natural next place for that requirement to land. If written policy is still sitting near half this time next year, the resolution will not come from corporate discipline. It will come from a lawsuit or a statute.
Sources
- Managers say they are using public AI tools to prepare for hard conversations (HR Dive, August 26, 2026)
- PI Study Finds CEOs Overestimate Manager Readiness as AI Changes Difficult Conversations (The Predictive Index / PR Newswire, August 2026)
- Stanford widened the young-worker gap to 19% and named the mechanism (LostJobs, August 12, 2026)
- Luna fired its first human. It lost its own rulebook first. (LostJobs, August 17, 2026)
- The most anxious workers aren’t avoiding AI. They’re the occasional users. (LostJobs, August 19, 2026)
