AI Coaching vs Human Coaching: What the Latest Research Reveals About Leadership

Every HR team in the Kingdom is now being pitched AI coaching. The offer is attractive: a conversational tool available at any hour, in Arabic or English, at a fraction of the cost of a credentialed coach, for every manager rather than the top fifty.

The question worth asking is not whether the technology works. It is what it works for, and for whom. Two randomized controlled trials have now tested exactly that, and they reached different conclusions about different groups of people. Read together, they are more useful to a Saudi HR team deciding where to put next year's budget than any vendor demonstration.

What the research actually found

The 2022 trial: parity, with students

Terblanche, Molyn, de Haan and Nilsson ran two longitudinal randomized controlled trials with undergraduate business students in the UK, published in PLOS ONE in 2022. One group received six one-hour sessions with professionally qualified coaches over six months. Another used Coach Vici, a goal-theory chatbot available around the clock. A total of 327 participants were analyzed across eight measurement points over ten months.

Both coached groups improved significantly more than the control groups, with almost identical effect sizes. As the authors put it, "the AI coach was as effective as human coaches at the end of the trials." The two coached groups never differed significantly from each other.

Two details inside that result matter. The human group reached a significant advantage over the control earlier, and the chatbot "trailed the human coaches slightly throughout the eight measurements." And within the AI group, the participants who used the chatbot more than the median of six sessions gained substantially more than those who used it less. Dose mattered.

The 2026 trial: a different answer, with managers

In February 2026, de Haan, Terblanche and Nowack published a three-arm randomized controlled trial in Human Resource Development International. This time the participants were not students. They were middle and senior managers at a global education company, drawn from around 34 countries, and the human coaching arm used 74 accredited executive coaches. Of 114 enrolled, 99 completed as intended.

The results separated. On goal attainment, human coaching produced significant improvement against the control on both goals measured, with moderate to large effect sizes. The AI group did not reach significance on either. On perceived stress, the human group improved significantly; the AI group did not. Both groups moved on general self-efficacy, though the authors note the AI effect did not hold up in regression. The abstract is blunt: the study "demonstrated substantial effectiveness in diverse coaching outcomes but only for human coaching."

The attrition figure is the one most HR teams should sit with. Fifteen participants did not complete as intended, and twelve of them were in the AI group. The authors describe AI attrition as roughly ten times higher than in the human arm.

The authors also disclose that one co-author develops the chatbot used and another has a commercial interest in one of the measures. That disclosure is a point in the study's favor rather than against it, and it is more than most vendor white papers offer.

Why the two studies disagree, and why that is the useful part

The obvious reading is that newer research overturned older research. The more accurate reading is that the two studies tested different populations on different problems.

A student working toward a defined personal goal over six months is well served by structure, prompts and accountability. That is most of what the goal-attainment work needs, and a chatbot can supply it at any hour.

A senior manager's hardest material is a different shape. It is a peer relationship that has quietly turned adversarial, a decision they have been postponing for three months, a doubt about a strategy they are publicly committed to, or the gap between how they believe they come across and how their team experiences them. That work depends on challenge, on reading what is not being said, and on a relationship with enough trust to risk being wrong out loud.

The attrition gap points to the same thing. People keep showing up for a person who is expecting them. Keeping a standing appointment with software, through a quarter when the project is late, asks for a kind of self-direction that stretched managers rarely have spare.

What this means for HR and L&D teams in Saudi Arabia

The Kingdom's own flagship leadership programs already answer this question in practice. Misk's Fellowship program lists coaching as one of five core elements, with monthly one-to-one sessions focused on self-awareness and long-term goals, and reports more than 3,469 coaching and mentoring sessions delivered. The 2030 Leaders program, built for accomplished Saudi leaders with at least ten years of experience and run with an international business school, supports participants with a coaching process alongside its modules and its national capstone project. Musk reports 615 alumni of that program to date.

Neither program replaced that human element with software, and neither treats coaching as a substitute for the taught curriculum. They pair the two, which is the structure the research supports.

Three features of the Saudi market make the distinction worth taking seriously rather than treating as a procurement detail.

• The pace is unusual. Leaders in new entities, giga-projects and newly formed regional headquarters are making consequential decisions earlier in their tenure than they would in a slower market. That is precisely the high-stakes, judgment-heavy territory where the 2026 trial found AI coaching did not move the needle.
• National talent development is personal. Developing Saudi nationals into senior roles involves questions of readiness, scope, visibility and stakeholder relationships that are specific to one person's situation. A curriculum can prepare a group. The step up is individual.
• Confidentiality carries weight. When the subject is a conflict with a senior stakeholder or doubt about a mandate, leaders need to know who can see what. That is a question about the coaching relationship and the provider's data handling, and it is answered differently when the other party is software.

A practical way to split the work

Rather than choosing one, decide what each is for.

• Reasonable uses for AI coaching tools, on current evidence: reflection prompts between sessions, goal tracking, preparation before a difficult conversation, practice at structuring feedback, and access for populations where one-to-one coaching will never be funded. Note the dose finding from the 2022 trial: the benefit appeared in people who used the tool consistently, so adoption support is not optional.
• Work that the evidence puts with a human coach: senior leaders in critical roles, anyone in transition, situations where the obstacle is a relationship or a pattern of behavior rather than a plan, and anything that requires being challenged rather than assisted.

It is worth being exact about what professional coaching is, because AI tools are sold under the same word. Coaching is not consulting, mentoring, training, or therapy. A coach does not supply answers or teach a syllabus. The work is structured reflection: examining assumptions, testing an approach in a real situation, and reviewing what happened. The 2026 results suggest that the challenge and relationship inside that process are doing real work, not decorating it.

Questions to ask a provider selling AI coaching

• What is the evidence, specifically? Ask for peer-reviewed research on a population like yours, not satisfaction scores.
• What does the tool do when someone needs challenging rather than supporting?
• What are the completion rates in organizations like ours, not the sign-up rates?
• Who holds the session data, where is it stored, and how does that sit with Saudi data protection requirements?
• Where does a credentialed human enter the process, and on what trigger?

On that last point, check what standard any provider holds its coaches to. Working with an ICF-certified coach means a defined competency framework and published ethics standards sit behind the engagement. A credible digital coaching platform uses technology for matching, scheduling, access and reporting, and keeps a qualified human in the coaching conversation itself.

What to expect, and what not to claim

No coaching format guarantees a business result, and that applies to both sides of this comparison. What the current evidence supports is narrower and more useful: with managers, human coaching produced measurable movement on goals and stress where an AI chatbot did not, and people stayed with it.

For a Saudi organization, reasonable expectations from coaching for managers delivered by credentialed coaches are leaders who can name what they are doing differently, decisions that were stuck getting made, and problems surfacing earlier. Treat any AI tool as support around that, and hold it to the same evidence standard you would apply to the human side.

Frequently asked questions

Is AI coaching worth anything at all? Yes, for the right task. The 2022 trial found a chatbot as effective as human coaches for goal attainment among students, and the benefit concentrated in people who used it consistently. Match it to structured goal work rather than to complex leadership material.

Will AI coaching replace human coaches? Current evidence does not support that for leadership populations. The 2026 trial with middle and senior managers found significant effects only for human coaching, and far higher dropout in the AI group.

Can we run both? That is the sensible structure for most organizations. Use AI tools for scale, prompts and practice, and credentialed coaches where the stakes, the seniority or the sensitivity are high.

Is AI coaching cheaper? Per seat, usually. Per outcome, compare completion rates and what each format is capable of addressing before treating the lower number as a saving.

What about confidentiality? Ask where session data is stored, who can access it, and how the provider handles personal data under Saudi requirements. Make your privacy team part of that conversation rather than relaying the answers secondhand.

Deciding where each belongs

If you are building a leadership development plan for the year and weighing an AI coaching tool against credentialed coaching, the research gives you a defensible way to split it rather than a reason to pick a side.

If you want to work through which of your leaders need human coaching and where technology genuinely helps, book a discovery call with CoachBase. We can look at your leadership population, where the pressure sits, and what a realistic mix looks like.