What Is AI Coaching and How Does It Work? A Guide for Educators

Every autumn, a careers tutor at a mid-sized university runs into the same wall. Two hundred final-year students need mock interview practice before graduate recruitment opens. She can give maybe thirty of them a proper practice run with substantive feedback. The other 170 get a handout and a wish of good luck. She knows what good practice looks like. She knows she could help all of them, but she simply cannot be in 200 rooms at once.

This is the problem AI coaching is being sold to solve. Before deciding whether it can, it helps to know what the term actually means, because a lot of very different products are wearing the same label.

So what is AI coaching?

AI coaching is the use of artificial intelligence to give learners personalised practice and feedback on a skill, usually through a chat, voice or virtual-character interface. For communication and soft skills, that typically means a learner rehearses something (an interview, a presentation, a difficult conversation), and the system responds with specific feedback on how they did.

It’s worth separating AI coaching from a couple of terms it gets muddled with. AI tutoring and intelligent tutoring systems are built to teach and test academic content, adjusting difficulty and offering hints on the actual learning materials, whether it be maths, coding or reading. Coaching works on something different: goals, reflection, self-awareness and behaviour change. A chatbot only becomes a “coach” once it’s designed around a coaching model and a defined development outcome.

There’s an honest debate about whether AI can truly “coach” at all. Graßmann and Schermuly (2021) described AI coaching as “machine-assisted” coaching, and a 2024 paper in Coaching: An International Journal of Theory, Research and Practice argues that, judged strictly, today’s tools don’t fully meet the definition of coaching, because coaching’s essence is a human developmental relationship. This is undoubtedly a valid criticism, and is often why AI coaching tools are combined with human-led intervention or discussion to provide that additional element.

The gap it’s stepping into

The reason AI coaching is growing has less to do with technology and more to do with a persistent shortfall in the skills that are hardest to teach at scale.

The World Economic Forum’s Future of Jobs Report 2025 found that employers expect 39% of workers’ core skills to change by 2030. Analytical thinking tops the list of sought-after skills, with roughly seven in ten companies calling it essential, closely followed by resilience, flexibility and leadership. These sit right alongside technical skills in employer demand.

The gap shows up on both sides of the recruitment desk. NACE’s Job Outlook 2025 reports that more than three-quarters of employers name communication as a critical attribute in new hires. Yet comparing NACE’s student survey with employer data reveals a stubborn perception gap: students consistently rate their own communication, leadership and professionalism higher than employers rate them, with the widest gaps running around 25 to 30%.

Cengage Group’s 2025 Graduate Employability Report puts a blunter number on it. Nearly half of graduates (48%) felt unprepared to apply for entry-level roles in their field, and only 30% landed a full-time job related to their degree.

Communication is learnable. What’s scarce is repeatable, low-stakes practice with feedback, delivered to every student. Career tutors can’t keep up with the demand. That scarcity is precisely where AI is being aimed.

The three types of AI coaching tools

“AI coaching” covers a spread of products that behave quite differently. Roughly, they fall into three groups.

1. Conversational coach: a text or voice chatbot that helps a learner set goals and reflect, often built on a GPT-style model wrapped around a coaching framework like GROW. Coach Vici, developed by Nicky Terblanche at Stellenbosch, is a well-studied example. Coaches like this are always available and completely non-judgemental, which lowers the barrier to trying again. Their weakness is that they can drift into generic advice and lack authentic human connection.
2. Roleplay-and-feedback simulator, and this is where communication skills tend to live. A learner practises a realistic scenario (a job interview, a patient conversation, a workplace disagreement) with a virtual character, then gets structured feedback on how they handled it. The appeal is safe rehearsal of high-pressure moments that have real world consequences. Having these vital skills prior to experiencing them in the workplace sets them up for success. This is the category Bodyswaps works in.
3. Blended human-plus-AI platform, where professional human coaches use AI for scheduling, session summaries and nudges between sessions. Research suggests AI can usefully support human coaches with the admin while the human keeps hold of the emotional and relational work.

Industry commentators sort the market along broadly similar lines. As Torch has described it, the market is sorting itself into distinct categories from programmatic chatbots through to AI-augmented human coaching. The label on the box matters less than which of these four things is actually inside it.

How AI coaching works, in plain terms

For a communication coach, the process behind the scenes runs in four steps.

First, it captures what the learner says. Speech is converted to text by automatic speech recognition. Second, it analyses that on two fronts at once: the content (is the answer structured, clear, relevant?) and the delivery (pace in words per minute, how many filler words, tone of voice).

Some of this is simply counted from the transcript; the rest is judged by the model against a rubric. These rubrics can cover a range of different competencies, using pre-existing frameworks to generate specific, usable feedback. These tools often flag filler words such as “um” and “uh,” measure speaking pace, and surface these back to the user. Third, it delivers that feedback through a dashboard, and conversational coaches may talk back using text-to-speech.

AI coaching works in tandem with roleplay, which adds another layer of impactful learning. Because a virtual character is playing the other person in the scene, the learner is judged on behaviour in context: did they listen actively, show empathy, keep a clear thread? That’s a closer match to real communication than assessing a monologue in isolation.

Does it actually work?

The evidence is encouraging, though it comes with clear limits.

The most-cited study is a pair of randomised controlled trials by Terblanche, Molyn, de Haan and Nilsson (2022) in PLOS One. Over ten months, both human coaches and the AI coach helped clients reach their goals significantly more than the control groups. What surprised the authors was that the AI coach was as effective as the human coaches on goal attainment by the end of the trials — a finding they described as unexpected given AI’s lack of true emotional intelligence.

On feedback specifically, Escalante, Pack and Barrett (2023), in the International Journal of Educational Technology in Higher Education, compared AI feedback with human tutor feedback for university language learners. They found no difference in learning outcomes between the two, and student preference split almost evenly. Their recommendation was a blended approach that uses the strengths of each.

For the simulation side, a 2024 systematic review and meta-analysis by Cho and Kim in Frontiers in Psychiatry pooled ten studies of around 1,000 nursing students and found VR simulation had a moderate, statistically significant effect on communication skills (Hedges’ g = 0.44). Critical thinking showed a larger effect again. The picture isn’t all positive: some secondary outcomes, including self-efficacy and confidence, weren’t significantly affected, and the studies varied a lot.

The pattern across the research is consistent. Narrow, goal-focused AI coaching can match human feedback on defined outcomes, and it comfortably beats giving learners no feedback at all. What it doesn’t replicate is the depth, trust and emotional read of a good human coach.

What good looks like

All of this points in one direction. AI is well suited to scale, and educators are well suited to depth.

Used well, AI coaching handles the part our careers tutor could never cover: giving all 200 students unlimited, private, judgement-free practice, whenever they want it. It also builds a clear picture of who is struggling with what, which frees her scarce time for the students who most need a real conversation.

For the people responsible for whole programmes, that same data has a second use. It shows, at cohort and institution level, whether communication skills are actually improving over time, which is the kind of evidence quality assurance and accreditation reviews increasingly expect.

This is the same logic behind deliberate practice, the learning-science idea that competency comes from repeated performance, specific feedback and gradual refinement. Ericsson’s own research noted that one-to-one instruction is usually capped at about an hour a week. AI’s real contribution is loosening that cap.

Good implementations also stay honest about the limits. Models trained on human data inherit human bias, so a career coach could quietly under-represent women as leaders unless it’s audited for fairness. Coaching involves sensitive personal data, so privacy and consent aren’t optional.

And there’s a real risk of over-reliance: researchers have started to describe a kind of “metacognitive laziness” when learners defer to AI too readily. The International Coaching Federation’s AI Coaching Framework (2024) and UNESCO’s guidance land in the same place. Keep a human in the loop, and treat AI as scaffolding around teaching rather than a replacement for it.

Additionally, the coach will only be as good as the learning design which underpins it. If it relies on the AI recalling knowledge from the web (or possibly inventing it if it can’t find a good source), it could lead to a generic and inconsistent experience that leaves learners with a lot to be desired. A carefully-crafted AI coach aligns with focused learning goals and gives everyone fair, well-informed feedback.

Where to start

If you’re weighing this up for a module, a whole programme, or a decision that reaches across your institution, a few practical steps to ensure your pilot runs smoothly.

1. Pick one use case where scale is the bottleneck. Mock interviews, presentation practice or difficult-conversation rehearsal are all good candidates, because they’re valuable, repeatable and challenging to deliver to a large cohort with personalised feedback.
2. Design for practice rather than novelty. Insist that the tool selected offers repeated attempts, feedback tied to a clear rubric, and visible progress over time.
3. Vet vendors on evidence, not marketing. Ask for peer-reviewed or independent results, fairness auditing and data-protection compliance. Be wary of headline confidence figures that come from the vendor’s own materials.
4. Finally, decide in advance what would change your mind. If students plainly don’t trust the AI feedback, move to a blended model. If skill gains plateau, add richer scenarios or human coaching. If an audit turns up a fairness or privacy problem, pause before scaling.

The short version

AI coaching won’t replace the careers tutor. What it can do is make sure that when graduate recruitment opens, all 200 of her students have practised, not just thirty. The technology is maturing, the evidence for focused and well-designed tools is solid, and the skills gap it addresses is real and well documented. The institutions getting value from it are the ones treating it as a way to give every learner the practice that used to be rationed, with educators still controlling the parts that matter most.

Further reading

See it for yourself: try Bodyswaps or book a demo with our team.

FAQ

What is AI coaching?

AI coaching is when a computer program helps you practise a skill and gives you feedback. For skills like speaking or interviews, you talk to the program and it tells you how you did and how to get better.

How does AI coaching work?

You speak or type. If you typed, the program compares what you’ve written to your goals. If you speak, the program turns your words into text, checks what you said and how you said it, and compares it to a set of goals. Then it gives you tips you can use next time.

Can AI coaching replace teachers?

No. AI is good at giving quick practice and feedback to lots of people. Teachers are better at the harder parts, like empathy, trust and judgement. The best setups use both.

Does AI coaching help students get better at communication?

Studies say yes, when the tool is well made. Students who practise with well-crafted activities and receive good AI feedback improve more than students who get no feedback. It works best as extra practice, not the only lesson.

References

1. World Economic Forum (2025). Future of Jobs Report 2025. weforum.org

2. National Association of Colleges and Employers (2024/2025). Job Outlook 2025 and Student Survey. naceweb.org

3. Cengage Group (2025). 2025 Graduate Employability Report. cengagegroup.com

4. Terblanche, N., Molyn, J., de Haan, E., & Nilsson, V. O. (2022). “Comparing artificial intelligence and human coaching goal attainment efficacy.” PLOS One. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0270255

5. Escalante, J., Pack, A., & Barrett, A. (2023). “AI-generated feedback on writing.” International Journal of Educational Technology in Higher Education. https://link.springer.com/article/10.1186/s41239-023-00425-2

6. Cho, Y., & Kim, M. (2024). “Effects of virtual reality simulation on nursing students’ communication: a systematic review and meta-analysis.” Frontiers in Psychiatry. https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2024.1351123/full

7. Graßmann, C., & Schermuly, C. C. (2021). “Coaching with artificial intelligence.” Journal of Leadership & Organizational Studies. https://www.frontiersin.org/journals/psychiatry/articles/10.3389/fpsyt.2024.1351123/full

8. Ericsson, K. A. (2015). “Acquisition and maintenance of medical expertise: deliberate practice.” Academic Medicine. https://academic.oup.com/academicmedicine/article-abstract/90/11/1471/8350635?redirectedFrom=fulltext

9. International Coaching Federation (2024). AI Coaching Framework and Standards. coachingfederation.org