This paper describes an integrated approach for detecting inconspicuous contents in text. Inconspicuous contents can be an opinion or goal that may be disguised in some way to mislead automated methods but keeps a clear message for humans (e.g., terrorist sites). Our methodology hypothesizes that patterns that convey inconspicuous contents can be extracted, represented, generalized, and matched in unknown text. The proposed approach is meant to complement data-intensive methods (e.g. clustering). Data-intensive methods are fast but are susceptible to variations in frequency, do not discern meaning, and require a large corpus for training. Our approach relies on manual engineering for natural language interpretation and pattern extraction using no more than ten examples, but is sufficiently fast to complement a real-time application.