Seventy-four percent of CEOs and business leaders believe their managers are confident handling tough employee conversations on their own. Only 20% of managers say they would rather prepare alone, without a framework, coaching guide, or input from HR. That gap comes from The Predictive Index's 2026 AI and People Leadership Survey, released August 18, based on 399 managers and 208 CEOs and business leaders. The data exposes a leadership blind spot with direct operational consequences: executives assume the management layer is ready, while managers are signaling they need backup before sitting down with someone whose pay, performance, or future at the company is on the table.
Forty-two percent of managers told Predictive Index they know what they want to say in a hard conversation. They just don't know how to say it. That's a delivery problem, not a confidence problem. It shows up in predictable places: 22.8% struggle most with delivering criticism constructively, 19.5% with anticipating how an employee will react, and 14.3% with staying objective once the conversation gets tense.
Managers are turning to public AI tools
According to HR Dive's August 21 report on the survey, 72% of managers said public AI tools are useful for preparing for difficult conversations. A manager can test an opening sentence, ask for a cleaner way to explain a performance issue, or rehearse how an employee might push back before the meeting starts. Few organizations teach managers how to tell someone the truth when the truth is going to hurt. AI for Management training addresses this gap directly, but adoption lags behind the need.
Anthony Belluccia, an I/O psychologist at The Predictive Index, drew a sharp distinction in HR Dive's report: trusting a manager and knowing they're set up for a specific hard conversation are two different things. Trust is not training.
The privacy risk most companies haven't addressed
The number that should stop an HR leader cold is 44%. That's the share of managers who told The Predictive Index they had entered an employee's name or performance details into a public AI tool. Not a vague scenario. Not a sanitized practice prompt. A real person's work record, carried into software the employer may not control.
Sixty-eight percent of leaders in the same survey said they're concerned about employee information being put into public AI platforms. Yet only 45% of organizations said they have a formal written policy for AI use in performance management. Most companies have the risk before they have the rule. A performance writeup, a pay dispute, or a termination conversation can become a legal problem fast, and HR Dive noted that experts have already warned employers about bias and legal ramifications when AI enters performance management. Once an employee's name and review details have been pasted into a third-party tool, the company has given away some control over information it was supposed to protect.
For startups, the danger is sharper. A 40-person company may not have a thick HR manual, but it almost certainly has managers using ChatGPT, Claude, Gemini, or whatever tool is already sitting in the browser. If those managers are handling feedback, compensation, or exits, the company needs a plain rule before the next hard conversation happens. AI for Human Resources courses increasingly cover governance frameworks for exactly this scenario.
Where AI helps and where it stops
None of this means the coaching itself is worthless. A manager who rehearses a hard conversation is more likely to be clear than one who walks in cold. AI can help with pacing. It can turn a rambling note into a sharper script. It can make a vague complaint sound more specific, which is often the difference between feedback an employee can act on and feedback they simply resent.
But the tool does not know the employee. Joel Pipher, VP of AI Transformation at The Predictive Index, said in the company's release that generic AI gives the same kind of answer to every manager because it lacks insight into the person on the other side of the table. That is the limit. A model can suggest language, but it cannot know whether the employee needs directness, time to process, context from a prior conversation, or a manager brave enough to stop reading from the script.
Why this matters for executives and strategy
The real story isn't that AI is bad at emotional conversations. It's that leadership has no idea how much of that emotional labor has already been outsourced. Managers are telling companies where they need help. Companies are answering with silence, loose access to public tools, and too few written rules. If you're responsible for organizational strategy, don't wait until the first messy termination to discover how your managers prepare. Give them a sanctioned place to rehearse, a rule about what employee data cannot leave the company, and enough training that they don't need to paste someone's name into a chatbot just to find the first sentence.
Your membership also unlocks:
