Generative modelling aims to accelerate the discovery of novel chemicals by directly proposing structures with desirable properties. Recently, score-based, or diffusion, generative models have significantly outperformed previous approaches. Key to their success is the close relationship between the score and physical force, allowing the use of powerful equivariant neural networks. However, the behaviour of the learnt score is not yet well understood. Here, we analyse the score by training an energy-based diffusion model for molecular generation. We find that during the generation the score resembles a restorative potential initially and a quantum-mechanical force at the end, exhibiting special properties in between that enable the building of large molecules. Building upon these insights, we present Similarity-based Molecular Generation (SiMGen), a new zero-shot molecular generation method. SiMGen combines a time-dependent similarity kernel with local many-body descriptors to generate molecules without any further training. Our approach allows shape control via point cloud priors. Importantly, it can also act as guidance for existing trained models, enabling fragment-biased generation. We also release an interactive web tool, ZnDraw, for online SiMGen generation ( https://zndraw.icp.uni-stuttgart.de ).