Quick summary
Stealth coaching is the pattern of using ChatGPT before hard conversations, after bad meetings, and ahead of negotiations, in place of a human coach. Millions of people already do this without calling it coaching.
Before a hard conversation with a manager, an employee opens ChatGPT and describes the situation. Before a big negotiation, a sales rep asks it to play the other side. After a meeting goes sideways, someone pastes in what was said and asks what they should have done differently. Before telling a direct report their performance is slipping, a manager asks ChatGPT to help them find the right words. Before asking for a raise, someone rehearses the pitch line by line. None of these people would call this coaching. That is exactly what it is.
What is stealth coaching?
Stealth coaching is the practice of using a general-purpose AI chatbot, most often ChatGPT, in place of a human coach. People use it to prepare for or process a real workplace situation, without labeling it coaching or telling anyone it happened. Stealth coaching is a term used in this article to name that pattern. It is not an established academic term, and no peer-reviewed research currently studies it under this label. It happens without a session being scheduled, a program being named, or anyone else finding out.
The behaviors themselves are specific and common. Rehearsing what to say before a tough conversation. Debriefing a meeting that did not go well. Practicing a negotiation before it happens for real. Drafting how to open a resignation conversation, or how to give a direct report critical feedback without it landing badly. Each of these matches an established coaching method covered in our coaching glossary. Conversational practice and the coaching feedback loop are the closest matches. The only difference is that no human coach or formal label is attached.
This is not a fringe habit. OpenAI’s own research analyzed 1.5 million conversations across ChatGPT’s 700 million weekly active users. It found that 49% of messages fall into a category the researchers call “Asking.” The report describes this as a growing category that shows people value ChatGPT most as an advisor, not only as a tool for finishing tasks. Stealth coaching is a specific, high-stakes slice of that broader advisor behavior.
Why nobody calls it coaching
Nobody calls this coaching because admitting it would mean admitting they needed help, and coaching still carries a lingering stigma of being in trouble. Ask any of these people what they were doing and almost none would say “I was coaching myself.” They would say they were “just checking something” or “getting my thoughts straight.” That gap between the behavior and the label is not an accident. It is a leftover from years of organizations using the word “coaching” as a euphemism for corrective action. Admitting out loud that a conversation needs rehearsal can feel like admitting a weakness. That feeling is especially strong for a manager who is supposed to already know how to manage people.
A chat window solves that problem by removing the audience. Nobody sees the draft messages, the false starts, or the version of the conversation that got deleted and rewritten three times. That privacy is exactly what makes the habit so widespread and so invisible at the same time. People are not avoiding coaching. They are avoiding the exposure of being seen needing it, and a chatbot lets them get the practice without the discomfort of asking a person for help.
Worth remembering
The real reason nobody names this habit is not reluctance to be coached. It is reluctance to be seen needing it.
Can you actually use ChatGPT for coaching?
Yes, with real limits, and the direct research on AI chatbot coaching is more mixed than either side of this debate tends to admit. ChatGPT can hold a coaching-style conversation reasonably well for a single session. It was not built as a coaching tool, but it can approximate one when prompted well.
A 2022 randomized controlled trial by Terblanche, Molyn, De Haan, and Nilsson tested a goal-focused AI chatbot coach against a no-coaching control group over ten months. The chatbot group’s goal attainment rose 55%, more than double the control group’s 24% rise. A 2026 randomized controlled trial by de Haan, Terblanche, and Nowack ran a more direct comparison. It tested human coaches, an AI chatbot, and a waitlist control side by side across 114 senior leaders. Human coaching produced significant improvements in goal attainment and other outcomes. The AI chatbot did not perform significantly better than the waitlist control in that trial.
Taken together, these two studies do not fully agree. AI chatbot coaching outperformed no coaching in the 2022 study. It did not clearly outperform no coaching in the more rigorous 2026 study, which also found human coaching pulling ahead of both. The honest summary is that AI chatbot coaching can help under some conditions. Whether it reliably matches structured human coaching, or even beats a general lack of coaching, is not yet settled.
Limitations of using ChatGPT as a coach
ChatGPT falls short as a coach in four specific ways. It has no reliable memory between sessions, no accountability for commitments made, generic advice with no organizational context, and a real privacy exposure. Whatever the research ultimately settles on, the structural gap is easy to see in practice. On a standard ChatGPT plan, there is no guaranteed memory of last week’s conversation. That rules out the coaching cadence that makes coaching work over time. Nobody is checking whether the commitment made at the end of a session actually happened, so there is no real accountability. The advice is also generic by default. ChatGPT does not know the company’s tools, a manager’s actual communication style, or the specific product a rep is about to pitch. Someone has to manually explain all of that context every single time.
There is a sharper risk too. In 2023, Samsung banned employee use of ChatGPT company-wide after engineers pasted confidential source code into it. Amazon, JPMorgan Chase, and several major banks put similar restrictions in place around the same time. The same risk applies to a coaching-style conversation. Pasting a confidential HR situation, a coworker’s name, or sensitive performance details into a consumer AI tool creates a real exposure. That tool was never built for workplace confidentiality, and most people are not thinking about that risk in the moment.
Here is what that gap looks like in a specific scenario. A manager needs to tell a direct report that their performance has slipped. They open ChatGPT, describe the situation in a few sentences, and get back three well-organized talking points and a suggested opening line. It reads well. They feel ready.
In the real conversation, the direct report brings up something that never came up in the rehearsal. Maybe it is a specific missed deadline, a disagreement about who owns a task, or a comment about feeling unsupported. The manager’s prepared opening line does not fit anymore, because ChatGPT never knew the history between them or how this particular person reacts under pressure. The manager improvises from there, which is exactly what they would have done with no preparation at all. The only difference is that this time, they walked in feeling falsely confident instead of appropriately nervous.
Worth remembering
Feeling ready and being ready are not the same thing. A rehearsal without real context prepares you for the conversation you expected, not the one you get.
The table below summarizes the structural gap between an unstructured stealth coaching session and a structured coaching practice.
| Dimension | Stealth coaching with ChatGPT | Structured coaching practice |
|---|---|---|
| Memory | No reliable memory between sessions by default | Retains goals and history across sessions |
| Cadence | Whenever the person happens to open the app | Runs on a set, recurring schedule |
| Accountability | Nobody checks whether the commitment happened | Follow-up on the previous session’s commitment is built in |
| Privacy | Confidential details entered into a consumer tool not built for it | Built for confidential workplace conversations |
| Organizational context | Generic, has to be manually explained each time | Aware of the company, role, and situation |
From stealth coaching to a real practice
The fact that millions of people already do this without being told to is a signal, not a quirk. It means people want to rehearse hard conversations and debrief bad ones. They are willing to do it on their own time, with whatever tool is closest at hand, and quietly enough that nobody else has to know. The gap is not motivation. The gap is structure, the memory, cadence, accountability, and privacy that a general-purpose chat window was never built to provide.
EasyCoach is an AI coaching product, built for exactly this kind of rehearsal. Full disclosure, Easygenerator, the company publishing this article, makes EasyCoach. It gives employees a private, structured space to prepare for a real conversation, whether that is a negotiation, a piece of difficult feedback, or a high-stakes pitch. Unlike a general chat tool, it remembers the goal, tracks progress across sessions, and stays inside a workplace-appropriate environment. If you already do this instinctively, EasyCoach gives that instinct real structure.
