Listeners rarely notice the structure of a radio hour, which is rather the point. A station that holds attention across a long drive is running a plan that decides what plays, in what order, and how often it comes back. The plan is old, it predates streaming by decades, and most of it still holds up when the transmitter is replaced by an HTTP stream.
The central tool is the hour clock, sometimes called the format clock or hot clock. It is drawn as a circle divided into wedges, each wedge representing a position in the hour and the type of content that belongs there rather than a specific song. A typical music clock places a recognisable song at the top of the hour to hold listeners through the junction, a news or traffic slot at a fixed minute so the audience learns when to expect it, advertising breaks at predictable positions, and station identification at set intervals. Because the clock specifies categories rather than titles, the same clock can run every hour of the day while the actual music changes underneath it.
Songs are sorted into rotation categories by familiarity and age. Currents are the newest and most heavily played, recurrents are recent songs on the way down, and gold or library tracks are older titles kept for recognition. Each category is assigned a rotation speed, and the arithmetic matters more than intuition suggests. If a category holds fourteen songs and the clock gives that category four plays per hour, a given song returns roughly every three and a half hours. Programmers call this turnover, and it is category size divided by plays per hour. Too fast and the station feels repetitive within a single commute. Too slow and the audience never builds familiarity with the songs the station wants to own.
Pacing works across the hour rather than within a single song. Tempo is mapped so that energy rises into a break and settles afterwards, keys and textures are varied so two similar sounding records do not sit back to back, and vocal density is spread out. Presenters are scheduled the same way. Talk segments are placed where the audience is most likely to be listening rather than wherever they happen to fall, which is why the strongest speech content on a music station usually sits in the first quarter of the hour.
Internet stations inherit most of this and drop some of it. Licensed terrestrial broadcasters in the United States must transmit a station identification at the top of each hour, a rule enforced by the Federal Communications Commission, and equivalent obligations exist under other national regulators. A purely online station is generally outside that particular requirement, though music licensing obligations still apply in full. What carries over is the discipline: playout software such as Liquidsoap or AzuraCast schedules by category and rotation rule in exactly the way a broadcast automation system does, because the underlying problem of holding an audience has not changed.
The practical takeaway for anyone running a small stream is that random shuffle is the weakest option available. A library of ten thousand songs played at random will repeat unpredictably, cluster similar records by accident, and give the listener no sense of where they are in the hour. Sorting the library into three or four categories and giving each a rotation rule takes an afternoon and produces a station that sounds deliberate.
References:
Keith, M. C. (2010). The Radio Station: Broadcast, Satellite and Internet (8th ed.). Focal Press.
Federal Communications Commission. Station Identification Requirements, 47 CFR 73.1201.
Berry, R. (2006). Will the iPod kill the radio star? Profiling podcasting as radio. Convergence, 12(2), 143-162.