Aims. Time-series spot modelling was used to follow the longitude changes of active regions responsible for the light variability of FK Com between 1987Com between −2004. Methods. The photometric data are analysed in the time-series mode of a spot modelling code. A scenario of one polar and two low-latitude active regions (hereafter spots, for simplicity) depicts the light variations very well. The role of the polar spot remains unclear because photometry in general does not provide direct latitudinal surface resolution, however, Doppler imaging results of FK Com also show very high latitude or even polar spots besides the low-latitude ones. We also used a light-curve inversion method to confirm some of the results. Results. The two low-latitude spots slowly migrate around 90• and 270• longitudes with quasiperiods of 5.8 and 5.2 years. The spots prefer to stay alternately on one or the other, but on the same hemisphere of the star, with a separation of typically 90−140• . We monitored a flip-flop in the light curve of FK Comae in 1999. The two low-latitude spots, being ≈140−180• from each other during the season, gradually decreased until they both practically vanished. Shortly thereafter, two new spots appeared and started to grow. One of the new spots was near the location of the old one, whereas the other turned up 90• shifted in longitude; consequently, the activity as a whole was shifted to the other hemisphere of the star. We followed a phase jump in 1997, when the two low-latitude spots got closer in longitude and finally merged, or else one of them vanished. A new spot appeared soon, shifted by 100• in longitude, but the activity remained on the same hemisphere. Conclusions. The difference between flip-flops and phase jumps is demonstrated. The derived longitude changes of activity centres may allow us to better constrain the theoretical modelling on the time-behaviour of stellar magnetic activity.