Teachable Newsfeed

01 / Problem

Personalised feeds adapt, but their logic stays hidden.

News apps rank content algorithmically while controls often sit behind menus or separate settings. Readers may not understand why a story appears, how to influence the feed or whether an action changed anything.

02 / Research

Study how people understand and correct recommendations.

03 / Core insight

Explain consequences, not the algorithm.

The interviews suggested that users did not need technical explanations of recommendation systems. They needed familiar controls, immediate feedback and observable changes in the feed.

04 / Exploration

Reject a memorable interaction when it adds too much work.

The early concept asked users to drag and group story cards spatially. It made the teaching metaphor tangible, but required too much precision on mobile and turned personalisation into a separate task that competed with reading.

05 / Iteration

Keep the metaphor, replace the interaction.

The final direction used continuous sliders with Less, Neutral and More zones. Repeated iterations added semantic colour, explicit labels, visible state, Reset and Apply actions, and confirmation after the feed reordered.

06 / Final model

Use two levels of control with one feedback behaviour.

07 / Reading flow

Keep control close without letting it dominate.

Preference setup is optional, the Home feed stays central and every personalisation surface remains one step from reading. Changes are reversible, and the same slider language appears in onboarding and recommendation settings.

08 / Gestalt

Use visual grouping to make system behaviour legible.

Proximity connects controls to affected content, common region binds each card, similarity creates predictable repeated structures and figure-ground gives temporary emphasis to confirmation.

09 / Reflection

A working model, with its limits stated.

The project translated research into interface decisions and a working prototype. Recommendation behaviour is simulated with a curated dataset, the participant group is limited and perceived control was not formally validated. A usability evaluation would be required to measure the model's effect.