Linking earthquake-triggered sedimentary imprints in lakes to ground motion parameters is essential for quantitative paleoseismology. However, current approaches rely on empirical ground motion prediction equations (GMPEs) and use a single time-averaged shear-wave velocity (V_s30) as a simplified site response proxy for a whole lake. We established a 3D shear velocity model to compute site-specific GMPE predictions and applied 2D numerical site response simulations for Lake Riñihue, Chile, to evaluate local ground motions for the 1960 M_w 9.5 Valdivia and 2010 M_w 8.8 Maule earthquakes. Even with site-specific V_s30 inputs, 2D simulations predict peak ground accelerations (PGA) and peak ground velocities (PGV) that exceed GMPE estimates by more than a factor of two. By stepwise modification of model properties and testing additional flat-layered reference models, we demonstrate that impedance contrasts between stratigraphic units influence overall ground motion amplification, whereas multi-scale basin geometry controls its spatial distribution, generating localized ground motion spikes. Earthquake shaking in lakes can produce surficial sediment remobilization (SSR) and soft-sediment deformation structures (SSDS) in-situ. Comparison of site-specific ground motions with sedimentary records from lake cores shows that SSR and SSDS are independent processes controlled by different ground motion components. SSR depth is primarily controlled by slope angle and PGV, with its patchy spatial occurrence reflecting frequency-dependent site response, whereas SSDS is controlled by PGA, with different thresholds for progressively increasing deformation types, while also predicts deformation thickness. Our findings highlight that site-specific ground motion reconstruction is essential to accurately link ground motion parameters to lacustrine sedimentary imprints.