Efficient extraction of pertinent terms from job advertisements (JAs) relies on feature engineering, specifically the selection of textual markers. Yet, current methods lack comprehensive assessment of markers’ quality in terms of their informative value and context-specific relevance. To address this issue, we introduce a fuzzy-inference-based framework, aimed at optimizing textual markers within JAs documents. This framework operates on three principles: ambiguity estimation of markers, assessment of the information quantity they convey, and their ongoing relevance evaluation within an organizational context. We demonstrated its effectiveness by applying it to 30 markers across 73 recruitment processes. The assessment resulted in the selection of five markers, improving the F1-score by 5% for relevant term extraction from JAs, via a possibilistic belief–desire–intention architecture. Six additional markers yielded a minor enhancement of results but significantly boosted model explainability. Thus, our framework significantly bolsters the extraction accuracy and clarity of the underlying model for JAs.