HR & Recruitment
Why AI is so quick to recommend the average in people decisions
And why HR should use it as a thinking partner – not as the judge.
That generative AI delivers plausible, professionally worded average answers is not a new phenomenon. For HR, this mechanism becomes particularly relevant as soon as a general recommendation is meant to turn into a decision on selection, development, potential, succession or reorganisation.
At that point, it is not enough for an answer to sound reasonable. People decisions set priorities, create opportunities and have consequences for real individuals. That is precisely why HR must scrutinise more closely which assumptions a recommendation rests on, which perspectives are missing, and who ultimately bears responsibility for the judgement.
The distinct HR contribution therefore does not lie in observing that AI can be generic. It lies in a robust usage logic: demand options instead of verdicts, bring in real organisational context, force counterarguments, and keep the responsibility for the decision and its justification with HR and leadership.
The average sounds remarkably professional
Generative AI is strong at condensing plausible patterns. Ask it for a good leadership profile, for transformation competencies, or for high-potential criteria, and you will often receive sensible, cleanly structured answers.
The problem: sensible is not automatically specific. Many recommendations are equally plausible for very different companies.
For HR this is risky, because our topics quickly become normative. What kind of leadership do we want? Which behaviour do we reward? What does potential mean at our company? What role does culture play? Which skills are genuinely critical to our strategy?
An eloquent AI answer can be technically correct and still wrong for your company.
In HR decisions, context matters more than elegance
A competency model that sounds sensible online does not necessarily fit the actual work. An interview guide can be formally clean and still overlook exactly the signals that matter for the role.
This applies especially to difficult decisions: selection, potential, development, reorganisation, succession or leadership. Here it is not only about patterns, but about consequences for people and the organisation.
That is why AI in HR should not take on the role of decision-maker. It can generate hypotheses, structure options, supply counterarguments and reveal blind spots. The judgement stays with people who know the strategy and the context.
Four rules for better HR use
Ask for options, not a verdict. Have several plausible paths developed, including risks and counterarguments.
Provide real company context: strategy, culture, roles, conflicting goals, past decisions and current framework conditions.
Have the AI attack its own recommendation. Ask under which conditions the proposal would be wrong, unfair or organisationally harmful.
Keep the decision and its justification with HR and leadership. Especially in decisions about people, a system recommendation must never blur the assignment of responsibility.
The decisive test
A simple test helps: would the same AI give almost the same advice to a different company with a similar question?
If so, that is not worthless. It can be a good starting point. But it is not yet a company-specific HR decision.
The added value only emerges when general patterns meet the organisation's own context and are checked, challenged and taken responsibility for by people.
What this means for HR competence
The better AI formulates its answers, the more important HR's ability becomes to tell good answers apart from merely good-sounding ones.
This shifts competence requirements. Prompting alone is not enough. We need more diagnostic thinking, better questions, an understanding of data and models – but just as much clarity about culture, organisation and responsibility.
The future strength of HR therefore does not lie in letting AI decide as much as possible. It lies in using AI in a way that makes our own judgement better.
AI is most valuable in HR when it expands our thinking – not when it replaces our judgement.