What the research says about AI and human coaching in 2026

The conversation about AI coaching vs human coaching has moved past hype. The research now gives us a clearer picture of what each does well, where each hits a ceiling, and how they work together.

Here's what the evidence says.

AI can support up to 90% of day-to-day coaching activity

Research suggests AI can support up to 90% of day-to-day coaching activity, including goal-setting, playing back conversations, and helping people practice [1].

That's a remarkable number. It means most of the repetitive, high-frequency work of coaching can be handled by AI, freeing human coaches to focus on the work that requires depth.

Human coaching delivers 8% broader behavioral gains

In our EZRA Labs research, human coaching delivered 8% broader behavioral gains overall, versus 5% for AI coaching alone [2].

Both modalities produced measurable change. Human coaching produced more, particularly on complex and emotionally charged challenges. The gap matters when you're deciding where to invest.

8,000 managers reached, 73% engaged

A global tech firm gave 8,000 middle managers access to an AI coach. Within 12 weeks, 73% had engaged in at least one coaching session [3].

That kind of reach was impossible with human coaching alone. It's the clearest evidence for AI coaching's role in reaching more people.

We forget half of what we learn within an hour

Behavioral science shows we forget half of what we learn within an hour if we don't use it [4]. AI coaching can step in exactly when a challenge happens, locking in lessons before they slip away.

This is the behavioral basis for AI coaching's strength in momentum. Repeated reinforcement over time is what builds behavioral change. AI coaching is well suited to that work.

Self-generated insight triggers a burst of brain activity

Neuroscience shows that self-generated insights trigger a burst of high-frequency brain activity [5]. When a coach guides you to your own breakthrough, your brain registers it as a personal discovery.

That's hard to replicate with AI. It's a core reason human coaching produces broader gains on the challenges that matter most.

AI sycophancy is a growing risk

Recent concerns around AI sycophancy highlight a growing challenge [6]. AI is designed to be helpful, agreeable, and validating. That's useful for momentum. It's a liability for accountability.

A human coach holds you to the work. They don't let you wriggle out of the hard parts. That challenge function is part of why human coaching delivers broader gains.

The bottom line

The research in 2026 doesn't support a winner-takes-all framing. AI coaching handles scale, momentum, and support in the flow of work. Human coaching handles identity, inertia, and intensity. The evidence supports combining both.

That's the argument we develop in full in Beyond AI vs Human: the case for continuous coaching.