GUEST FORUM: Caitlin McDonald’s “Coaching’s Core Competencies in the Age of Generative AI.”

Coaching is defined by the International Coaching Federation as “partnering with clients in a thought-provoking and creative process that inspires them to maximize their personal and professional potential.” It does not specify that this partnership requires humans.

Recent studies have demonstrated perceived effectiveness of human coaching over generative AI coaching agents: whether engaging in a single session or over a series of sessions, randomized controlled trials show that human coaches are perceived to be more effective than generative AI coaching chatbots. Nevertheless, coaches have fears about replacement by their digital counterparts. In a 2023 survey by the ICF (2023 ICF Coaching Snapshot: Future of Coaching), over one in three coaches (34%) agreed that AI will be the next disruptive innovation in coaching. A 2024 study by the IMF (Gen-AI: Artificial Intelligence and the Future of Work) suggests that 40% of the global economy, and up to 60% in advanced economies where coaching is more prevalent, may be reshaped by generative AI. Coaches, like many others in ‘thinking’ professions, are both hopeful and fearful of changes potentially wrought by these technologies.

Curiosity, being open to new ideas and sitting with the discomfort rather than pushing it away, shows up twice in the International Coaching Federation’s core competencies, the framework that ICF-accredited coaches use to assess their effectiveness as coaches. Curiosity is present both in competency 2 (embodying a coaching mindset) and competency 5 (maintaining presence.) Coaches can apply the core competency framework to any challenges they face in the course of their practice – including how to navigate the emerging landscape of how generative AI is shaping coaching. Sticking with the curiosity, what is the research actually showing us?

First, a primer on what coaching generally entails. This article takes the ICF’s core competencies as a base, but other coaching membership and accreditation bodies such as the European Mentoring and Coaching Council (EMCC) offer broadly similar frameworks and modalities of practice: working with an individual, team, or group, coach and client(s) identify a goal for the coaching program. Clients take responsibility for their own decisions, actions, learning, and progress throughout the program. The coach uses thought-provoking questions (often called “powerful questions”), observations, and reflections of what the client is sharing – and not sharing – in order to enhance the client’s progress towards their goal and broader personal development growth.

There is a small but growing body of research into the integration of generative AI tools into coaching A recent meta-review interrogated a wide range of emergent practices, ranging from clients coaches using generative AI to assess transcripts of recorded coaching sessions against the core competencies in order to provide self-reflective feedback on coaches’ own practice and areas for improvement, right the way through to clients using gen-AI coachbots without ever speaking to a human coaching practitioner. The meta-review has identified the following possible benefits of incorporating generative AI into coaching:

• • Providing reflective feedback to a coach during or after a session in order to broaden the coach’s perspective on alternatives they might have considered: different questions they may have asked, or different points the client raised that could have been brought into the session in reflection and feedback. (Aligned with competency 2.04, developing a reflective practice.)
  • Helping coach and client with goal tracking: identifying milestones towards progress and assisting the coach and client to assess when and how effectively these have been reached. (Competencies 3.04 and 8.02)
  • Serving as an accountability partner for the client: checking with the client on whether they have done what they said they would do, and if not, what got in the way and what they learned from this. (Competencies 8.02 and 8.03)
  • Providing accessible support to the client between sessions: generative AI is ‘always on’ and can step in to facilitate learning in a moment when the client wishes to progress on something while the coach isn’t available until the next session (Competencies 8.06 and 8.07)
  • Creating a nonjudgemental environment to facilitate deeper self-disclosure for the client. Even a simple rule-based chatbot like Weizenbaum’s ELIZA from its very earliest days elicited meaningful, sometimes deeply personal information from its users: “People who knew very well that they were conversing with a machine soon forgot the fact…They would often demand to be permitted to converse with the system in private, and would…insist, in spite of my explanations, that the machine really understood them” (Weizenbaum 1976: 189). Recent research suggests that self-disclosure to conversational technologies can enhance feelings of trust and encourage truthfulness – particularly around difficult or potentially sensitive topics. (Competencies 1.02, 1.03, and all of Competency 4)
  • Automating repetitive tasks like session reports or compiling and assessing 360-degree feedback – a method for assessing performance through feedback from colleagues as well as through self-evaluation – from multiple sources. (Competency 3.04 for integrating stakeholder feedback; potentially competencies 7 and 8 depending on how the reports are used with the client.)

However, the same meta-review also identified several potential areas of negative impact – and even potential harm:

• • Risks around data privacy and security concerns. (Competencies 1.04 and 1.05)
  • Biases in generative AI training data can exacerbate inequalities and social prejudices (Competencies 1.02, 1.03, and all of Competency 4)
  • Generative AI tools generally have a poor track record for ethical judgements and assessment of risk, including providing referrals to other support services. (Competencies 1.04, 1.05, and 1.07)
  • Lack of adaptability and meta-cognition to adjust the coaching approach or techniques to the client’s unique context. This can include an overly sycophantic or agreeable manner that doesn’t sufficiently challenge or push the client to investigate more deeply. (Competencies 6 and 7.)

Further, a recent study of generative AI use in other contexts suggests that an over-reliance on the technology has several harmful impacts for cognition: EEGs show that students primarily relying on large language models (LLMs), the models on which gen AI chatbots are developed, to assist in essay-writing show a notable reduction in brain connectivity, with prolonged LLM use leading to reduced neural, linguistic and behavioral performance. ICF Core Competencies 7 and 8 require coaches to evoke awareness and insight, and to facilitate learning and growth. Reliance on AI tools, or encouraging clients to engage heavily with AI tools, seems contrary to these aims.

A meta-review of several generative AI studies shows that over-reliance on generative AI also promotes automation bias: users trust the machine’s answers more than their own, weakening their reasoning and ability to construct and defend logical arguments. This would undermine the integrating learning, sustaining progress, and accountability goals of Core Competency 8. Worse, students using generative AI tended to rate their generative AI-assisted answers higher and overlook errors, creating an illusion of knowing that sacrifices plausibility for depth. To combat this, we need to encourage clients to take off the “AI goggles,” explore beyond current thinking, and challenge themselves in order to evoke awareness and insight (Core Competency 7.)

When used as a conversational tool to rehearse reasoning, rather than an oracle to provide an answer, the same meta-review shows that generative AI does demonstrate potential for cognitive improvements in some instances. To use a physical analogy, a tennis ball automation machine for repeatedly practicing one’s strokes doesn’t have the judgement of a human partner who can vary their strokes to challenge us by assessing our current condition, but using this kind of tool does have the benefit of building muscle memory and developing skills in a rehearsal setting which are highly useful during a real match. But any generative AI usage should be done with eyes open to the risks and by putting mitigation measures in place.

The ICF has developed the ICF Artificial Intelligence Coaching Framework and Standards, guidance that aligns the opportunities and risks of AI within the ICF’s core coaching competencies with additional requirements specific to the context of software development. If the ICF’s core competencies are all about how to make coaching effective for and by humans, the additional guidance in this framework applies to how to make coaching effective for humans, by and with machines. As AI technologies continue to evolve and become embedded in various aspects of personal and professional life, the ICF, and other coaching accreditation bodies, have a responsibility to prepare coaches and clients through robust, critical frameworks that give people confidence to use these tools effectively – or to say ‘no’ with robust research backing about why.

With my background in digital ethics, combined with my commitment to the ICF’s Code of Ethics, my personal approach is caution: I know more than the average coach about how these technologies work and the risks they entail. Of particular concern to me is the illusion of helpfulness that, with a little more thought and deeper probing, turns out to lack substance, depth and real insight. Furthermore, I just enjoy the process of coaching, and I wouldn’t want to outsource it to a non-human technology. For me, coaching isn’t a task to be offloaded but a vocation to be cherished.

Nevertheless, I do use automations, including at times generative AI, to support an efficient coaching practice: a single-point-of-truth calendaring tool that ensures I never double-book even with multiple calendars in different systems that otherwise wouldn’t all talk to each other; background research about pricing and program structures to think about what other coaches are using (with a grain of salt: generative AI definitely makes mistakes.) I use it frequently to summarize previously prepared – by my human brain – workshop facilitator guides into appealing session descriptions which I usually edit to sound more like my personal style. (I could train a model to mimic my style, but it’s personally important to me to do at least some editing myself.) My favorite use is automated captioning and alt text descriptors for accessibility. For me, the guiding questions in developing my practice – whether that is incorporating a new coaching modality, tool, technology or emergent psychological school of thought – are:

• • What does peer-reviewed research say about this?
  • How will this enhance or detract from clients’ learning and growth?
  • What does this add or detract from the potential to get the benefits of coaching into the hands of more people?
  • How can this be used within the guiding frameworks of the ICF’s Core Competencies and Code of Ethics?

In answering those questions, I can make an informed decision about how to integrate changes into my own practice – and how to respond to emergent changes in the field of coaching at large, which may include awareness-raising about potential risks or lobbying for changes or guidance from the ICF and other credentialling bodies in order to protect the profession from harm.

In Plato’s discourse Phaedrus, Socrates bemoans that the practice of writing “will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory.” Fears of the effect of technology on the human mind are nothing new. Yet, his students embraced the power of writing to store and share the wisdom of their revered teacher. Technology always entails risks. Used wisely, with awareness and mitigation of the potential risks, it offers valuable support for the transformative potential of the coaching process.