AI coaching has been shown to be moderately effective, with research suggesting that AI chatbot coaching is very good (and cost effective) at providing a neutral, non-judgemental space in practice. But the average employee is probably already wary about privacy and confidentiality when using AI at work. And in my research, many people were less likely to use an AI coach if it were provided by their employer – so I feel like it’s already quite a hard sell unless it has clear evidence of benefits and confidentiality.
Lately I’ve seen more than one AI coaching platform provider talk about using coaching data for organisation level behavioural / productivity metrics as part of their product’s value proposition. From an employee point of view, that seems like an additional red flag, and makes coaching sound like another channel for the organisation to monitor, interpret or score them rather than a genuine safe space for reflection and development.
There’s also a possible legal and compliance issue using coaching data in this way. Under the EU AI Act, AI uses in employment and management is treated as particularly for high-risk for regulation. So AI products marketed around behavioural inference, or employee-level evaluation could create serious compliance and admin questions, extra implementation work and potentially significant cost in an EU context (see EU AI Act Annex III (4)). It’s already prohibited to use AI systems to infer emotions from biometric data (facial images, voice, keystroke dynamics), in the workplace or educational since February 2025 (AI Act Article 5 1(f))
It’s interesting to compare the ethics of AI and human coaching in organisations. Internal and external human coaches also have to manage confidentiality, sponsor relationships, record-keeping, and legal obligations. But in good human coaching, confidentiality is usually treated as an essential boundary condition of the work, and the conversation is held within the limits of human judgement and memory rather than turned into a scalable data asset. AI coaching seems to blur that boundary somewhat, as the commercial promise can depend on turning the conversation itself into analysable organisational data.
AI coaching platform providers may already be thinking carefully about these challenges, and the recently delayed timetable for some high-risk obligations under the EU AI Act (until December 2027) may give organisations more time to prepare. But this extended deadline should not mean a free-for-all for high-risk AI use. The Act reflects a basic concern about AI in employment contexts: it could too easily turn reflection and development into data extraction and assessment. If AI coaching is to have a credible future in organisations, it needs an ethics-first design principle: coaching data should not become a backdoor route into employee surveillance, otherwise it may erode the trust in coaching itself