Effective implementation of sustainable development policies often stalls at the community level, where success depends on identifying credible local champions. This study proposes and validates a behavioural-classification mechanism—the Altruism–Neutrality–Selfishness (ANS) experiment—that isolates individuals whose intrinsic motives align with long-horizon ecological goals. A four-set, 20-round ANS game was administered to 167 volunteers across Delhi, Jaipur, Meerut and Bijnor. In each round participants chose to (A) donate to a public “society fund,” (N) abstain, or (S) extract resources; Sets 1–2 were played anonymously to reveal private preferences, whereas Sets 3–4 were played with public scrutiny to test reputation-driven costly signalling. We analyzed cumulative choices using a robust cut-off rule based on the sample median plus one median absolute deviation, producing five mutually exclusive classes: Altruist, Pseudo-altruist, Selfish, Indecisive and Fluctuating. Non-parametric Friedman and Kruskal–Wallis tests confirmed statistically significant differences both within classes (card preferences, p < 0.05) and between classes (behavioural profiles, p < 10−9). Only 12% of participants met the stringent criteria for “true” altruism—consistent contributions even when unobserved—whereas 58% exhibited context-dependent, fluctuating behaviour. The ANS taxonomy delivers three actionable outcomes. First, it provides programme managers with a reliable screening tool for assigning leadership and monitoring roles to genuine altruists rather than reputation-seekers. Second, knowledge of the sizeable fluctuating segment guides the design of norm-based nudges that altruistic leaders can activate. Third, the experiment’s low cost and statistical robustness make it scalable for municipalities, firms and NGOs seeking to embed behavioural insights into grassroots sustainability initiatives.