Variability in gene expression causes genetically identical cells to exhibit different phenotypes. One probable cause of this variability is transcriptional bursting, where the synthesis of RNA molecules randomly alternates with periods of silence in the transfer of genetic information. Yet, the molecular mechanisms behind this variability remain unclear. Experiments indicate that multiple biochemical states might be involved in the production of RNA molecules. Stimulated by these observations, we developed a theoretical framework to investigate the mechanisms of transcriptional bursting. It is based on a multistate stochastic approach that provides a full quantitative description of the dynamic properties in the system. We found that the degree of stochastic fluctuations during transcription directly correlates with the number of biochemical states. This explains experimentally observed variability and fluctuations in the quantities of the produced RNA molecules. The procedure to estimate the number of relevant biochemical states participating in the transcription is outlined and applied for analysis of experimental results. We also developed a general dynamic phase diagram for the transcription process. The presented theoretical method clarifies physical-chemical aspects of the transcriptional bursting and presents a minimal chemical-kinetic description of the process.