How Do Streaming Services Decide What to Make and Why Does So Much of It Feel the Same

[StoneCold]

Is it all algorithms all the way down the race to the bottom?

[CobyOlaleye]

The content decision process at major streaming services in 2026 is driven by data at a scale that would have been impossible for traditional broadcasters. Netflix, Max, Disney Plus and their competitors know not just what people watch but when they stop watching, which scenes cause skip-forward behaviour, which episodes cause cancellation, what thumbnail images produce the highest click-through rates, and how viewing patterns correlate with subscriber retention. This data is used to optimise content decisions.

The problem is that data-driven optimisation tends toward the centre of existing preferences rather than toward genuinely new things. Data tells you what people who are already subscribers have watched. It tells you less about what might attract new subscribers or what audiences would discover they love if they encountered it. The creative breakthrough that redefines what television can be does not usually look like anything in the existing training data. This is why established genres feel overdeveloped while genuinely experimental work is increasingly rare on major platforms.

The other structural factor is the metrics used to evaluate success. A show with 80 million views in its opening weekend is easier to justify than a show with 4 million views that generates intense critical conversation, awards attention, and subscriber loyalty that is hard to attribute directly. The first number is clean and computable. The second number's value to the platform is real but harder to demonstrate to shareholders.

The exception tends to be the flagship prestige project, the show a platform leads with to establish identity and generate awards attention. These get more latitude because their function is partially brand signalling rather than pure viewership. The House of the Dragon commissions, the Severance renewals, the Shogun productions, these exist partly to signal that the platform makes television worth taking seriously. The volume content around them follows the data.

[GhostRider89]

The thumbnail optimisation point is something most viewers do not realise. The image you see for a show when browsing Netflix is often not a representative still from the show but a specifically A/B tested image that produced the highest click-through rate. The show itself may look nothing like it

[MickFoley00]

The distinction between the prestige flagship and the volume content is visible once you understand it. The flagship is the show the platform promotes at award season and leads with in industry conversations. The volume content is what fills the catalogue and keeps subscribers subscribed between flagships

[Megan95]

The cancellation problem, shows being cancelled after one season before the story concludes, is partly a data problem. A show that builds slowly and pays off in a third season is invisible to metrics that are evaluated quarterly

[BiscuitTin]

International co-productions have sometimes escaped the data optimisation trap by serving multiple markets with less overlap in the test data. Something that performs well in South Korea and surprisingly well in Latin America might not have been predicted by US subscription data

[BigDog_Fan]

The binge release model changed the data available to platforms. When entire seasons are released at once you can see completion rates, where people stop, and how many episodes get watched in the first weekend. This data is then used to commission more of what those patterns suggested worked

[Cole75]

Data optimisation producing genre saturation is the central critique of streaming and it is valid. If data says crime dramas retain subscribers and generate completions, you commission more crime dramas. Eventually the crime drama market is saturated and the subscribers who were retained by novelty are retained by nothing