ARTIFICIAL INTELLIGENCE IN AGRICULTURAL EXTENSION COMMUNICATION: A SYSTEMATIC LITERATURE REVIEW OF 278 STUDIES

Abstract

Artificial intelligence (AI) technologies are increasingly deployed to modernise agricultural extension communication, yet the evidence base for their effectiveness remains fragmented and methodologically immature. This systematic review, conducted in accordance with PRISMA 2020 guidelines, synthesises evidence from 278 studies retrieved across five databases, covering the period 2015–2026, to characterise the landscape of AI enabled extension tools, quantify evidence certainty for key farmer outcomes, and identify critical methodological gaps. Of 1,156 retrieved records, 278 met inclusion criteria following deduplication and two-stage relevance screening; a subset of 30 full-text accessible studies received detailed quality appraisal using the Mixed Methods Appraisal Tool (MMAT) 2018. System development studies predominated (39.6%), with India (24.8%) and Sub Saharan Africa (18.0%) as the primary geographic contexts; chatbots (32.7%) and mobile/SMS platforms (29.5%) were the most common AI communication modalities. Only one randomised controlled trial (0.4%) was identified, and all 30 MMAT-appraised studies were rated ‘Not assessable’, reflecting the dominance of proof-of-concept designs over evaluative research. Applying a GRADE-adapted narrative grading approach, evidence certainty for knowledge gain and technology adoption outcomes was graded ‘Very Low’, while productivity and economic outcome evidence was insufficient for grading. These findings indicate that, despite rapid technological growth, the field lacks the experimental rigour necessary to inform large-scale policy commitments; future research must prioritise randomised evaluations, longitudinal cohort studies, equity- disaggregated outcome reporting, and offline-capable inclusive system designs to translate AI innovation into verified agricultural development impact.