Geophysical inverse problems are inherently ill-posed, non-unique, and nonlinear, posing substantial challenges for deterministic optimization methods that are prone to entrapment in local minima. Over the past three decades, metaheuristic optimization algorithms have emerged as powerful alternatives for solving complex geophysical inversion problems across multiple subdisciplines. This paper presents a comprehensive systematic review, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, of metaheuristic algorithm applications in geophysics from 1990 to 2025. A total of 312 peer-reviewed publications were identified from Scopus, Web of Science, Society of Exploration Geophysicists (SEG) Library, GeoRef, and Google Scholar databases, of which 186 met the inclusion criteria for detailed analysis. The review encompasses six primary geophysical domains: seismic data processing and inversion, gravity and magnetic field modeling, Electromagnetic (EM) and resistivity methods, seismological applications, well log analysis and petrophysics, and geophysical survey design optimization. Comparative performance analyses reveal that Particle Swarm Optimization (PSO) and Differential Evolution (DE) consistently demonstrate superior convergence properties and solution accuracy across most geophysical inversion problems, while Simulated Annealing (SA) maintains advantages for high-dimensional parameter spaces with rugged objective function landscapes. Hybrid approaches coupling metaheuristics with local gradient-based methods show improvements of 15–45% in misfit reduction with 30–60% fewer forward model evaluations. Bibliometric analysis indicates exponential growth in publications since 2010, with recent trends emphasizing Graphics Processing Unit (GPU)-accelerated implementations, deep learning surrogate models, and multi-physics joint inversion frameworks. The review identifies critical challenges including computational scalability for three-dimensional models, uncertainty quantification, and the gap between synthetic benchmarks and field data validation. Future directions including quantum-inspired metaheuristics, physics-informed neural network surrogates, and cloud-based distributed inversion architectures are discussed.