Research: How AI Helped Executives Improve Communication

Leading like a coach is a powerful way that executives can help employees to reach goals and improve their performance. Successful leader-coaches draw on strong communication skills, and are marked by their ability to ask good questions, resolve conflicts, and provide meaningful feedback to their direct reports. Yet not all executives know how to practice the skills required to be an effective coach and communicator.

Having worked with numerous executives on their coaching abilities, we know that one of the best ways to hone these skills is by having a human trainer observe executives’ conversations with others and provide feedback. Given the increasing capability of large language models (LLMs) to have “conversations” with humans, we wondered: could LLM-powered tools become effective sparring partners to sharpen executive coaching capabilities?

To test our theory, we created a tool, powered by GPT-4, to provide feedback to executives, like a human coach would. We tested our tool with 167 global executives from a wide range of industries. We found that AI-based coaching interventions can be an efficient, accessible, and potent way to improve essential leadership capabilities. And AI tools are available to almost everyone.

Two Methods of Using AI to Have Better Conversations—About Having Better Conversations

There are two primary ways that executives can interact with these tools to improve their communication and coaching skills: by asking for advice or asking for assessment.

In the first, executives can treat the AI like a leadership coach, and ask it for help in preparing for or analyzing difficult conversations. They can also prompt the tool to role play, and then explore how different answers may trigger different reactions.

Consider, for example, an executive who feels uncertain about the follow up in a situation with a team member where a fast reaction is warranted. A helpful first step would be to write a short, 250-word description of the situation. This process alone creates a moment of reflection that can help provide clarity. The written text can then be fed into the tool or the executive can simply speak to the tool, explaining the situation and then prompting it to act as a coach. This can be done with open questions like “What would you advise, as a coach?” or “How can I deal with this situation in a productive way?” The executive could also employ more specific questions, such as “Which communication style would you recommend in this case?” or “How do I apply a supportive communication style in a succinct manner?” or “How should I follow up?” The more specific the prompt, the higher the likelihood of a useful answer.

Another approach would be to have AI tools listen in to a conversation between an executive and someone else. After the conversation, the executive can ask the tool to assess their communication style. Again, the more specific the question, the better the outcome. Including a request in the prompt to provide specific examples delivers useful—and at times surprising—insights. Likewise, asking whether the executive’s word choice was appropriate or asking the AI tool “to assess whether the dynamics were productive” makes good use of the tool’s granular attention for the entire duration of the conversation. If the AI tool is fine-tuned to recognize a specific  framework, executives can ask for the “adequate choice of communication style” for the situation at hand.

How to Use AI to Become a Better Coach

In our experiments, we used AI to listen to and assess actual conversations, as we believed this would help executives better understand and develop their own coaching style. We used a version of GPT-4, fine-tuned with a set of pre-existing coaching conversations that we analyzed according to John Heron’s coaching styles framework. Heron’s framework categorizes interactions into six types: prescriptive (offering advice or suggesting solutions), informative (sharing knowledge or insights), catalytic (encouraging self-discovery and problem-solving), cathartic (sounding out the emotional state), supportive (providing emotional encouragement), and confronting (challenging assumptions or behaviors).

We used the tool in a variety of settings in which executives coached each other, drawing on real situations from their careers. Our tool “listened” to these conversations. After the session, the executive acting as the coach asked the tool about their coaching approach with specific questions such as:

• What is my coaching style according to Heron’s framework?
• Which styles do I use most—and is that effective?
• How effectively did I perform as a coach?
• What went particularly well in the conversation?
• What patterns do you see in how I ask questions?
• How could I improve my coaching skills?
• Do I tend to dominate the conversation, and if so, how?
• What do you see about my coaching style that you think I don’t see?

As a reference point and to critically assess the feedback given by the AI tool, human observers also watched the conversations and documented what they saw.

The Expected, the Unexpected, and the Surprising

In its analysis of the coaching sessions, the tool recognized the correct Heron coaching styles precisely. It also provided feedback and recommendations for each style’s situationally appropriate use. We also found that the AI-generated feedback and recommendations were in alignment with the findings made by the human observers.

As one senior executive from a large Scandinavian company put it: “Initially skeptical, I soon shifted gears and recognized the importance of understanding how to measure certain outcomes of our leadership. Utilizing technology can significantly enhance our leadership skills.” Another executive, from a Japanese company, reported: “[I] learned that AI can do a lot more than just crunch statistic data in an LLM.” An executive from a large Moroccan enterprise shared that, “Using this tool was incredibly practical and engaging. It offered an interactive and personalized approach, almost like having a dedicated coach readily available with tailored advice.”

Overall, the executives in our experiments found the AI-generated assessments largely useful, in some cases even surprising. In one instance, an executive noted having considered another approach but deciding against it. The tool detected this unexpressed choice and highlighted the other approach as an alternative path in its feedback report—a revelation that we had not anticipated.

Not unexpectedly, the tool proved most helpful in response to more specific prompts. The prompt “Can you give me specific examples?” yielded the most valuable feedback, according to the executives.

How Executives Rated the AI Companion’s Assessments 

The experiences of the executives that used the tool can be better understood when ordered within a framework we created. We call this framework NU, and it measures how Novel and how Useful the tool’s assessments proved to our participants.

NU’s first dimension measures the level of agreement or discrepancy between the perception that the executives had about their coaching style and the perception reported by the tool. Its second dimension measures how useful the participants found the feedback generated by the tool. This creates a matrix with four quadrants:

See more HBR charts in Data & Visuals

1. Zone of Validation: Feedback aligned with participants’ expectations and reinforced what they already knew.

About 30% of the executives we worked with fell into the zone of validation. Being in this quadrant reflects a state of reassurance and confirmation. Here, participants recognized the feedback as mirroring their own understanding of their coaching style. While this can be comforting, it also presents a potential limitation: a risk of complacency. For those in this zone, the challenge lies in moving beyond validation and identifying subtler opportunities for growth that may not yet be apparent.

2. Zone of Learning: Feedback was both surprising and useful, sparking new insights.

Approximately 55% landed in the zone of learning. This quadrant represents the most dynamic and transformative space. Here, participants experienced insights that challenged their assumptions and opened new pathways for development. They stated that they found the “conclusions suitable and new.” Also, they appreciated that the AI tool had a “granular attention” for the entire duration of the conversation and picked up on the tone of the speakers (providing feedback such as, “the question was too aggressive”).

For individuals in this quadrant, the experience highlighted how AI can surface blind spots and stimulate meaningful growth. This quadrant underscores the value of both curiosity and a growth mind-set in leadership.

3. Zone of Irritation: Participants found the AI tool’s insights  off-target or unhelpful.

Another 10% found themselves in the zone of irritation. Several of these participants reported that they missed more practical behavioral guidance. This quadrant signals a friction point, which often arises when feedback conflicts with self-perception or lacks immediate relevance. While irritation can be uncomfortable, it also presents us with an opportunity for reflection. We encouraged the participants in this zone to ask questions: What was it about the feedback that caused discomfort? Was there a kernel of truth that can be unpacked? What would be your approach to changing your behavior? Irritation, while challenging, can be a precursor to growth if approached with an open mind and a willingness to explore underlying causes.

4. Zone of Indifference: Participants were neither surprised nor found the analysis useful.

Only 5% of participants landed in the zone of indifference. For these individuals, the feedback might have been too general, too obvious, or misaligned with their goals. Indifference can signal a missed opportunity if it leads to disengagement. To shift out of this zone, participants might benefit from clarifying their expectations, asking more specific questions, or seeking alternative perspectives to deepen the relevance of the feedback obtained. All of which offers an opportunity to fine-tune the tool even more.

Not Perfect, but a Helpful Sparring Partner

While AI-powered coaching tools offer promising insights, there are important points that executives should consider.

First, to deliver their insights AI tools require access to conversations, which raises questions about data security and confidentiality. The sensitive nature of these conversations might deter participants from fully engaging unless robust data protection measures and clear consent processes are in place.

Second, AI occasionally produces inaccurate or misleading feedback (“hallucinations”) that risks misguiding development. Leaders must critically evaluate AI-generated insights and consider them alongside personal observations or peer input to confirm their validity. For this reason, we recommend executives that engage in this type of interaction to anchor their prompts in specific frameworks (such as Heron’s). These frameworks give an anchoring point for the assessments generated by the AI “companion” and enable executives to easily identify instances in which the tool may be hallucinating.

Third, those willing to experiment with this type of tool must also be aware that AI struggles to interpret the cultural nuances, emotional dynamics, and unspoken cues, which are often critical in the coaching environment. These contextual gaps highlight the need to use AI tools as complements, not replacements, for human expertise.

Key Takeaways for Executives Who Want to Grow as Leaders

1. Embrace structured frameworks and tailor your prompts.

To fully leverage AI-based coaching, we recommend employing structured frameworks, such as Heron’s coaching styles, and asking precise, contextualized questions. The quality of the feedback delivered by the tool improves significantly when prompts are explicit and specific, for example, “Which aspects of my questioning style encourage self-discovery?” or “Can you share with me an approach to employing a ‘confrontation’ style that would help me address underperformance effectively?” By grounding queries in clear frameworks and narrowing their focus, executives can gain insights that are not only descriptive but also actionable, while reducing the risk of getting merely superficial answers. This structured approach deepens the coaching conversation and helps executives better understand where they need to shift their communication patterns.

2. Adopt a reflective stance to navigate unexpected insights.

Executives should be prepared for feedback that challenges their self-perceptions. Often, the most valuable discoveries can be found where one’s own intuition and the tool’s perspective differ. When the tool surfaces unfamiliar or uncomfortable observations, executives should resist the impulse to dismiss them. Instead, they should reflect on these surprises, ask follow-up questions, and consider experimental changes to their communication approach. Unexpected insights can catalyze meaningful development. Interestingly, many executives reported that it somehow feels easier to receive this type of assessment from an impersonal AI companion than from a fellow human. This suggests that AI companions can be “safe spaces” in which executives can ask questions that they don’t wish to ask their peers.

3. Combine AI feedback with a human perspective.

AI assessments often prove accurate and rich in detail, but these tools are most powerful when paired with human interpretation. Executives who discussed their AI companion’s feedback with human coaches or peers deepened their understanding and contextualized the insights gained. This dual feedback loop—AI-driven analysis complemented by a trusted, human sounding board—creates a more nuanced growth environment. By integrating machine and human input, executives reinforce the validity of certain findings, challenge unfounded conclusions, and translate broad suggestions into culturally and situationally appropriate actions.

4. Continually refine the process for sustained growth.

The process of improving communication through AI coaching is iterative. Early sessions may feel basic or only partially relevant, placing participants in the zones of validation or even indifference. Over time, as executives refine their prompts, incorporate new scenarios, and integrate previous learnings, the AI-powered feedback grows more meaningful and personalized. This iterative approach encourages leaders to maintain a growth mind-set over the long term. Rather than viewing the AI tool as a one-time diagnostic, they should see it as an evolving resource—one that, when consistently engaged with, helps them adapt, experiment, and master the subtle art of powerful communication.

A Final Piece of Advice: How to Prompt

In addition to the larger takeaways, our research highlighted three particular ways that executives used prompts to generate useful insights. We share them with you as helpful jumping-off points:

• Ask the AI tool to rewrite your statements in radically different emotional tones—playful, empathetic, assertive—and see which style resonates best.
• Ask the tool to flag overused phrases and filler words and suggest quick swaps that will sharpen your messaging.
• Start a conversation in one cultural context, then ask the tool to reinterpret it in another. This can help uncover hidden assumptions and strengthen global adaptability.

• • •

AI tools are transforming how leaders refine their coaching abilities. In a world where leadership conversations shape organizations, AI offers an innovative way to ensure those conversations are as impactful as possible. While these tools are not a substitution for human coaches—with their rich understanding of human connection, cultural context, and empathy—for leaders willing to experiment and reflect, these tools offer vast learning potential.