BACKGROUND: In Ethiopia, 38% of children under 5 years of age are stunted (low height for age). A novel government-led intervention called the Sustainable Undernutrition Reduction in Ethiopia (SURE) aims to tackle the burden of stunting by improving complementary feeding and dietary diversity among young children. The SURE programme design applies a transtheoretical model of behaviour change, whereby exposure to recommended infant and young child feeding (IYCF) and nutrition-sensitive agriculture messages is a first stage to adopting key behaviours. This qualitative study explored the fidelity and dose of the IYCF and nutrition-sensitive agriculture messages delivered by extension workers. METHODS: A qualitative study was conducted across four regions in Ethiopia (Oromiya, Amhara, SNNP and Tigray) between April and October 2017. Across the four regions, 81 key informant interviews, 90 FGDs and 81 observations were conducted with 180 extension workers, 18 development agents and 54 mother-father pairs. Digitally recorded audio files were transcribed verbatim, and the data were analysed based on a framework analysis approach using NVivo (version 12) by coding and categorising texts into major themes and sub-themes. RESULTS: SURE target households had the intended exposure to messages about exclusive breastfeeding, timing of initiation of complementary feeding, food groups, diversified food consumption, irrigation, rearing small animals and vegetables. Few households reported receiving messages on the content or frequency of complementary feeding of a child beyond 6 months of age. Frequency of household visits and hence exposure to SURE messages was also variable. Agricultural messages delivered during household visits focussed on improving standard agricultural practices and rarely covered the importance of nutrition-sensitive agriculture to improve household or child nutrition. CONCLUSION: Despite variability observed in the breadth and depth of messages delivered, large-scale behaviour change communication programmes can achieve moderate to good message exposure among target groups. Qualitative data provide an in-depth insight into fidelity and may supplement our understanding of programme roll-out and implementation. Further research is required to understand longer-term message saturation including frequency and reach.