Starch Selection in Sweet Food Systems: A Texture-Driven Framework Across Confectionery and Dessert Matrices

Sweet food systems present a uniquely complex environment for starch functionality across confections, baked goods, dairy desserts, fruit-based products, and bakery fillings. The role of starch varies fundamentally between product categories; therefore, its selection and application cannot be treated as generic. The sugar-dominated matrices of these products alter starch gelatinization temperature, retrogradation tendency, and water distribution in ways that vary with sweetener identity, concentration, and molecular structure; however, real sweet food systems are inherently multicomponent, making starch behavior in these matrices substantially more complex than model-system predictions suggest. This review provides an integrated starch selection framework for sweet food formulation, analyzing functionality and modification requirements across seven major product categories: caramel and toffee, gummies and jelly candy, hard candy, cookies and biscuits, bakery fillings, jams and fruit preserves, and custards and dairy desserts. The central principle is that starch selection must be matched to the dominant failure mode of the specific matrix because no single starch type or modification approach addresses the full range of functional requirements across sweet food systems. The sugar-starch interaction is established as the mechanistic foundation for product-specific selection, with particular attention to how reformulation pressures, including sugar reduction, fat reduction, and clean-label demands, alter this interaction in ways that make conventional reference data unreliable. Recommendations are explicitly characterized as directly evidenced by product-specific studies or analytically synthesized where direct evidence is absent, transparently identifying where experimental validation remains needed. Three cross-cutting formulation challenges are critically examined: the clean-label paradox, starch-flavor binding, and the limitations of predictive models in multi-ingredient sweet food matrices.