Background/Question/Methods: How can we understand the natural world if we cannot predict the dynamics of species interactions? A growing number of studies on AM fungal-plant-herbivore interactions show these interactions occur but are context dependent. To address which factors most strongly influence AM fungal-plant-herbivore interaction dynamics, I present a predictive framework based on a machine learning approach analyzing multiple studies in two AM fungal-plant-herbivore interaction systems from my group. I applied a machine learning approach to analyze data from both AM fungal-plant-herbivore interactions systems. Machine learning uses a permutation approach to identify which variables have the greatest influence on system dynamics (in this case which variables limit predictability). I used the randomForest program in R version 3.4.4, and ran 1000 permutations of a decision tree regression to come up with composite trees most closely predicting total plant or insect biomass from 10 variables. Results/Conclusions: The analysis revealed that the experiment the data was collected from, the plant genotype, and the plant source material had the greatest influence on plant biomass in AM fungal–plant–herbivore interaction systems. By contrast, neither plant nor herbivore species influenced plant biomass. The factors with the greatest influence on insect biomass in AM fungal-plant-herbivore interaction systems were plant genotype, plant species, presence of an endosymbiont, and herbivore genotype. The results suggest that in AM fungal-plant-herbivore interaction systems genotypic level effects will have the greatest modifying influence on plant biomass, while both species and genotypic level effects will modify the influence on insect biomass. These results should guide the selection of variables to test in future studies of AM fungal-plant-herbivore interactions.