Agent at Work Phase 2: Age-Coded Reasoning in UK Job Adverts
An analysis of how an AI system interprets age-related signals in job adverts.
Description
This report presents Phase 2 of the Agents at Work research series, examining how an AI system interprets age-related signals in recruitment language.
Building on Phase 1, which identified where age-adjacent language appears, this phase focuses on how those signals are understood and categorised by the system.
Using a structured evaluation approach, the report analyses how the model reasons about language, identifying recurring patterns in how age-related cues are interpreted across roles and contexts.
Rather than focusing on final classifications alone, the analysis examines the reasoning behind those judgements.
What This Report Does
Phase 2 examines:
- how age-related signals are interpreted by an AI system
- recurring categories of age-coded language
- how different types of cues appear across job roles
- how language is framed as youth-oriented, experience-based, or exclusionary
The report introduces a structured taxonomy of age-related cues derived from model reasoning.
What This Report Does Not Do
This report does not:
- assess real-world hiring outcomes
- determine employer intent
- provide compliance or legal judgements
- measure discrimination in practice
The analysis focuses on interpretation of language rather than outcomes.
Who This Is For
This report is intended for:
- researchers studying language bias and AI interpretation
- HR and recruitment professionals
- policymakers and governance specialists
- practitioners working with AI-supported evaluation systems
Research Context
This report forms Phase 2 of the Agents at Work series.
It builds on Phase 1 by moving from detection of signals to interpretation of how those signals are understood, providing a foundation for later phases examining system behaviour under repeated evaluation.
Why This Matters
Age-related signals in recruitment language are often implicit.
Understanding how these signals are interpreted helps clarify how AI systems may frame suitability, and how subtle differences in language can influence evaluation outcomes.
Licence and Usage
© 2025 Imogen Hull – Beyond the Average
Licensed under Creative Commons CC BY-NC-ND 4.0.
The underlying methodology, agent design and analytical framework remain proprietary.
