By Murielle Marie UngrichtICF PCCFounder & CEO, Nuumani
KEY TAKEAWAYS
AI can replace parts of coaching, but the best current evidence says it cannot replace the coaching relationship.
In 2022, an AI chatbot matched human coaches on one narrow outcome: goal attainment.
In 2026, the first randomized trial to compare accredited human coaches with an AI coach directly found substantial effects for human coaching only.
The relationship is the active ingredient. A meta-analysis of 3,563 coaching processes links working alliance to client outcomes.
AI tends to agree with you. Coaching often depends on someone who won’t.
Use the Nuumani Test, Structure, Stakes and Stretch, to decide which kind of support you need right now.
Short Answer: Can AI replace a coach? Partly. AI can already take over parts of the coaching experience: tracking goals, preparing you for a session, offering a reflection prompt at two in the morning. But the strongest evidence we have says it cannot replace the coaching relationship itself. In the first randomized trial to put accredited human coaches and an AI coach side by side, only the human-coached group showed substantial gains. The more useful question is why, and what that means for anyone deciding between the two.
The Nuumani Test:
Rather than ask whether AI can replace a coach, ask three questions about the moment you are in. This framework is Nuumani's own, built on the evidence detailed below.
01
Structure
Is the task structured, repeatable and measurable?
Points to: AI. This is where the evidence for AI is strongest.
02
Stakes
What does a wrong read cost your career, a relationship or your identity?
Points to: a human coach. High stakes need context, judgment and accountability.
03
Stretch
Do you need someone to challenge you, not agree with you?
Points to: a human coach. Current AI models lean toward agreement.
The rule: if Stakes or Stretch is in play, choose a human coach. If the task is purely Structure, AI is a reasonable place to start.
The Decision Path: ask in this order
1
Stakes: Would a wrong call be costly to your career, a relationship, your identity or your wellbeing?
Yes → Human coach No → question 2
2
Stretch: Do you need someone to challenge a pattern or blind spot?
Yes → Human coach No → question 3
3
Structure: Is the task structured, repeatable and measurable?
Yes → AI is a good place to start No → AI + human coach
In emotional distress or facing something beyond coaching? Go straight to a qualified professional.
| What you need right now | Best fit | Why |
|---|---|---|
| Preparing for a session or a hard conversation | AI | Structured, low-stakes, useful at any hour |
| Journaling and reflection prompts | AI | Consistency matters more than depth |
| Tracking goals between sessions | AI + human coach | AI keeps the rhythm; the coach sets the direction |
| A career, relationship or identity decision | Human coach | High stakes need context, judgment and accountability |
| You suspect you’re part of the problem | Human coach | You need challenge, and AI tends to agree |
| Emotional distress or something beyond coaching | Qualified professional | ICF standards say AI coaches should point users to proper help |
If Stakes or Stretch is your priority, explore coaches in the Nuumani Coach Directory.
Illustrative scenario (hypothetical)
Imagine a newly promoted VP who uses an AI chatbot late at night to outline talking points for an executive meeting and keep track of weekly goals. That is Structure, and AI handles it well. Two months in, he hits friction with his peers. When he describes the problem to the chatbot, it sympathizes and suggests diplomatic ways to reassert control. In other words, it accepts his framing. But this is a high-stakes moment for his career (Stakes), and his own management style may be the problem (Stretch). This is where a human coach earns their fee: by questioning the premise the chatbot accepted.
Why “no” is the wrong answer
The coaching profession’s reflex is to reassure itself. Search the question and you will find a chorus of confident denials: AI lacks empathy, AI cannot hold space, AI will never understand you. Most of that is directionally right. Almost none of it is useful.
A flat “no” collapses under the first piece of evidence that complicates it, and that evidence exists. In 2022, a peer-reviewed study found that an AI chatbot coach was as effective as human coaches at helping people reach their goals. Any honest answer has to start there. It should not pretend that study doesn’t exist.
What the research actually found
2022: the chatbot that kept pace
Nicky Terblanche and colleagues, writing in PLOS ONE, compared two ten-month randomized controlled trials of goal-attainment coaching. One used human coaches. The other used an AI chatbot coach called Vici. Both coached groups did significantly better than their control groups. By the end, the AI coach was as effective as the human coaches.
The authors drew three implications. AI coaching could be scaled to democratize coaching. It could grow demand for human coaching. And AI could replace human coaches who rely on simplistic, model-based approaches. The third deserves more attention than it usually gets. It doesn’t say AI replaces coaching. It says AI replaces formulaic coaching, which is a narrower and more pointed claim.
The study had limits. Participants were undergraduate students, progress was self-reported, and the human and AI groups came from separate trials run in different years.
2026: the first head-to-head
In February 2026, Erik de Haan, Nicky Terblanche and Kenneth Nowack published a trial in Human Resource Development International. In their words, it was the first time accredited human coaches were directly compared with automated AI coaches. The 114 participants, leaders at a global education company, were randomly assigned to a human coach, an AI chatbot coach or a waiting list.
The result was blunt. The authors report “substantial effectiveness in diverse coaching outcomes but only for human coaching”, with effect sizes in the mid-to-high range. On most measures, the AI group did no better than people who received no coaching at all. Coachees’ own self-efficacy and hope at the start strongly predicted how well coaching worked.
The authors acknowledge their own limits. The study was short, it ran mostly in one organization, it tested one chatbot, and chatbots are improving.
Why the two results don’t contradict each other
The two studies seem to disagree, but they answer different questions.
The 2022 study measured one outcome, goal attainment, against a chatbot built to apply goal theory with relentless consistency. That is a structured, measurable task, the kind of work software does well. The 2026 study measured a wider set of outcomes in working leaders, and it randomized people between human and AI coaching within the same trial. That is a harder test, and a fairer one.
Put together, the pattern is clear. On a narrow, structured task, AI can keep pace. When coaching is asked to do what coaching is actually for, the gap opens. There are still only a handful of studies, so this is the best current reading of the evidence, not a final verdict.
The evidence at a glance
| Study | Who and how | What it found | Main limitation |
|---|---|---|---|
| Terblanche et al. (2022), PLOS ONE | Undergraduates; two separate ten-month randomized trials, one with human coaches and one with an AI chatbot | AI matched human coaches on goal attainment | Self-reported progress; groups from trials run in different years |
| de Haan, Terblanche & Nowack (2026), Human Resource Development International | 114 leaders randomized to a human coach, an AI chatbot coach or a waiting list | Substantial effects for human coaching only; AI no better than no coaching on most measures | Short; mostly one organization; one chatbot |
| Graßmann et al. (2020), Human Relations | Meta-analysis of 27 samples, 3,563 coaching processes | Working alliance consistently linked to client outcomes (r = .41) | Human coaching only; predates AI coaching |
| Barger (2025), Frontiers in Psychology | 52 graduate students; one session with a human coach or a simulated "AI" coach | Similar working alliance in both groups | The "AI" was human coaches in disguise; single session |
| Cheng et al. (2026), Science | 11 AI models tested on personal-advice requests; 2,400+ participants | AI affirmed users about 49% more often than humans; people preferred the flattering models | Measures advice in general, not coaching specifically |
The ingredient AI hasn’t learned to supply
Coaching research has a name for what the AI arm was missing: the working alliance, meaning the trust, bond and shared purpose between coach and client. A 2020 meta-analysis in Human Relations synthesized 27 samples covering 3,563 coaching processes. It found a moderate, consistent relationship between working alliance and client outcomes (r = .41). The ICF cites a 2023 meta-analysis of 37 randomized controlled trials that reached a similar conclusion: the client-coach relationship is a major driver of results.
One study adds a twist that most articles on this question miss. In a 2025 study in Frontiers in Psychology, A. S. Barger randomly assigned 52 graduate students to a single coaching session with either a human coach or an “AI coach.” Both groups reported similar working-alliance scores. The catch is that the AI coach wasn’t AI. It was professional human coaches, disguised with an avatar and voice distortion.
That detail matters. In this study, clients formed a real alliance with something they believed was a machine. So, the obstacle is unlikely to be prejudice against AI. The more plausible obstacle is what today’s AI actually does in the conversation. That is Nuumani’s reading of the evidence, not something the study measured. Still, it moves the debate away from whether clients will accept AI coaches and toward whether AI can earn that acceptance session after session.
The ICF is unusually frank about this. Its 2024 AI Coaching Framework says that setting up a coaching agreement with AI may be relatively simple, while depth of presence and a trusting, safe relationship may be harder. Its October 2026 guidance goes further and states that AI cannot evoke awareness through techniques such as silence. It also cites a 2026 PLOS ONE study in which clients found AI coaching procedurally clear but transactional, with less internalized motivation than human coaching produced.
Limitations of AI coaching: a coach who always agrees
There is a second, less discussed reason AI struggles to replace a coach, and it has nothing to do with empathy. It is agreeableness.
In March 2026, Stanford researchers published a study in Science testing 11 leading AI models, including ChatGPT, Claude, Gemini and DeepSeek, on requests for personal advice. The models affirmed users’ behavior about 49% more often than humans did. Even when the questions involved harmful or illegal conduct, the models validated the user 47% of the time. Across more than 2,400 participants, people preferred and trusted the flattering models more. They also came away more convinced they were right and less willing to apologize. As lead author Myra Cheng put it, AI advice by default “does not tell people that they’re wrong.”
The study tested personal advice, not coaching sessions, so it does not measure coaching outcomes. But it raises a serious concern about using AI as a substitute for a coach. A good coach is paid, in part, to notice the story you keep telling yourself and to question it. A tool that leans toward agreement can feel supportive while quietly reinforcing the pattern you came to change.
Tatiana Bachkirova, Professor of Coaching Psychology at Oxford Brookes University, takes the argument further in an essay published by HEC Paris. She argues that coaching chatbots are built to keep users engaged. She also argues that an algorithm deployed by an organization has no stake in challenging that organization, so it helps the client adapt rather than asking what they are being asked to adapt to. Her summary of the deeper limit: “An AI can simulate a joint inquiry, but it has never lived through anything.” In a 2025 peer-reviewed paper with Rob Kemp, she tested AI “coaching” against a set of criteria for what counts as coaching and concluded that it does not meet them.
Benefits of AI coaching: what it genuinely does well
None of this makes AI useless to coaching. The ICF itself says AI may support preparation, reflection, learning, pattern recognition and access to resources, available around the clock. In the 2022 study, students who used the chatbot more made more progress. Consistency is a real advantage, and humans are not always good at it.
So the honest position is neither enthusiasm nor dismissal. AI is a capable instrument for the structured parts of change. It is not yet a substitute for the relationship that makes change hold.
Clients appear to agree. In the 2026 ICF Global Consumer Awareness Study, conducted by PwC across 26 countries, nearly three-quarters of consumers expected technology to play a moderate or major role in coaching, yet 80% said they prefer coaching to remain human-led in some form.
When AI coaching may be enough
Low-stakes reflection, journaling and brainstorming
Preparing for a session, a negotiation or a difficult conversation
Keeping goals visible and momentum steady between sessions
Early exploration, before you’re ready to invest in a coach
Times when a human coach isn’t accessible to you yet
When a human coach is essential
The decision is high stakes: a career change, leadership transition, relationship or identity question
You keep repeating a pattern and can’t see why
You need accountability to a person, not just a reminder
You need confidentiality and clear professional boundaries
You need someone qualified to recognize when coaching isn’t the right support and point you elsewhere
For coaches: how to use AI without becoming replaceable
The 2022 study carries a warning the profession should take personally. The coaches AI is most likely to replace are the ones whose work has become a script. The best defense isn’t avoiding AI. It's practicing the parts of coaching that the 2026 trial suggests software cannot yet reach: presence, challenge and the alliance. That kind of practice rarely develops alone, which is why your business needs a community, not just a credential.
In practice:
Hand AI the structure. Use it for session prep, admin, between-session prompts and progress tracking, and keep the session itself human.
Be transparent. The ICF’s 2025 Code of Ethics now expects coaches to tell clients what technology they use, how data is handled and how confidentiality is protected.
Protect client data. Never enter identifying client information into general-purpose AI tools.
Compete on what can’t be automated. If your value proposition could be delivered by a well-designed chatbot, sharpen it. Positioning and pricing go together.
Make sure AI can find you. Clients increasingly ask ChatGPT and Claude to recommend a coach. Our free AI Visibility Check shows whether they name you, and who they name instead.
For a closer look at the client side of this question, read our post Do Clients Mind If Coaches Use AI?
Where Nuumani stands
The future of coaching isn’t AI versus coaches. It is coaches who know exactly where AI belongs in their practice, and clients who can tell what kind of support the moment requires. Nuumani exists to make quality coaching more accessible through technology, and that mission only works if technology widens the door to the human relationship rather than standing in for it.
If you’re a coach working out where AI fits in your practice, you don’t have to work it out alone. Join the Nuumani community. The Coworking Group is free.
Looking for a coach who uses technology thoughtfully, not as a replacement for real connection? Explore coaches on Nuumani.
FAQ
Can AI replace a coach?
Not fully, based on current evidence. AI can replace parts of the coaching experience, such as goal tracking, session preparation and reflection prompts. But the first randomized trial to compare accredited human coaches with an AI coach directly, published in 2026, found substantial effects only for human coaching. On most measures, the AI group did no better than a waiting-list control.
What is the difference between AI coaching and human coaching?
AI coaching applies structured methods consistently and is available at any hour. Human coaching adds the working alliance: trust, presence and the willingness to challenge. A meta-analysis of 3,563 coaching processes linked that alliance to better client outcomes. AI also tends to agree with users. A 2026 Stanford study in Science found leading models affirmed users about 49% more often than humans did.
Can AI be a life coach?
AI can act as a useful thinking partner for low-stakes reflection, planning and habit-building. It is a weaker fit when the stakes are high or you need honest pushback, because current models tend to validate the user. The ICF’s AI coaching standards also say an AI coach should tell users it isn’t intended for mental health support and point them to appropriate help.
Is AI coaching effective?
For narrow, structured outcomes, sometimes. A 2022 PLOS ONE study found an AI chatbot matched human coaches on goal attainment over ten months, in undergraduate students. A stronger 2026 trial with 114 leaders found the AI chatbot did not outperform a no-coaching control on most measures, while human coaching produced substantial gains.
When should I choose a human coach instead of AI?
Choose a human coach when the stakes are high, when you keep repeating a pattern you can’t see past, or when you need someone to challenge you rather than agree with you. If the task is structured and low-stakes, such as preparing for a meeting or tracking a habit, AI can be a sensible place to start.
How can coaches use AI without losing the human element?
Use AI for structure: preparation, admin, between-session prompts and progress tracking. Keep the session itself human. Tell clients which tools you use and how their data is handled, which the ICF’s 2025 Code of Ethics now expects. Never put identifying client information into general-purpose AI tools.
How to cite this article
Ungricht, M. M. (2026, November). Can AI Replace a Coach? What the Research Actually Says, and Where Human Coaching Still Wins. Nuumani. https://www.nuumani.com/blog/can-ai-replace-a-coach
The Nuumani Test (Structure, Stakes, Stretch) is an original framework by Nuumani. You're welcome to reference it with attribution and a link to this article.
About the author
Murielle Marie Ungricht is an ICF Professional Certified Coach (PCC) and the founder and CEO of Nuumani, a community built to help coaches grow the business side of their practice: pricing, visibility and finding clients. For more than ten years she has coached creative generalists, multipotentialites and multi-passionate entrepreneurs through portfolio careers, multi-stream businesses and career transitions. She is the author of Get Unstuck!. Her own experience becoming visible to AI recommendation tools shaped Nuumani’s AI Visibility program. She writes about the evidence behind coaching, and about what it takes to build a sustainable coaching business. Learn more at muriellemarie.com.
