Reconstituting an elementary gene expression system inside self-assembled lipid vesicles to mimic the cellular synthesis machinery is at the core of the development of a minimal cell following a bottom-up synthetic biology approach. The ability to operate the expression of multiple genes in a controlled manner and to generate the output proteins with predictable dynamics in liposomes relies on the link between genotype and phenotype. Here, we established this link in surface-tethered liposomes producing proteins from a linear DNA template using a reconstituted transcription/translation/aminoacylation apparatus fuelled by external supply of feedstock. The amounts of entrapped DNA molecules and synthesized proteins were visualized by fluorescence confocal microscopy in individual vesicles. We showed that there exists no linear correlation between the amount of encapsulated genes and the level of output proteins, which is a consequence of the compositional heterogeneity between liposomes due to the low-copy number of some constituents, as well as interfacing differences with the nutrient-containing environment. In order to decouple gene activity from those sources of variability and, thus, infer the probabilistic occupancy of transcriptionally active genes in protein synthesizing liposomes, we developed a dual gene expression assay consisting of the production of two fluorescent reporter proteins of distinguishable colors from two different DNA templates. The stochastic color-coding of the vesicles was analyzed and compared to the color pattern expected from a Poisson distribution of encapsulated genes. Unexpectedly, we found that the apparent number of transcriptionally active DNA molecules in liposomes corresponds only to ca. 10% of the bulk concentration. We believe that our study provides new insights about the relationship between the genotype and phenotype in protein synthesizing liposomes, which is of primary importance toward the construction of a programmable artificial cell implemented with regulatory gene networks of predictable dynamics.