Generative artificial intelligence (GenAI) is gaining traction in higher-education pedagogy, yet evidence-based pathways for its deployment remain sparse. Text-mining 47 domestic cases, we identified three inter-locking routes—paradigm, activity and assessment—each moving from instrumental delivery to collaborative participation. GenAI enters the classroom as a co-participant, forming a teacher–AI–student synergy that risks cognitive dependency, shallow learning, integrity breaches and diminished social dialogue. Embedded in crowd-intelligence scripts and human–AI co-creation, it mediates collective meaning-making; in assessment, it powers real-time, multimodal co-evaluation across cognitive, behavioural and affective domains. Viewed through sociocultural theory, GenAI is a cultural tool whose internalisation hinges on social interaction. Safe, effective integration demands: (i) a teacher-led, AI-as-scaffold, student-centred governance model to mitigate risks; (ii) goal-oriented tasks, multimodal resources and blended spaces to deepen application; and (iii) a national–departmental–student co-assessment framework fuelled by multi-source data.