Designing a Shared Language for Recommendations

Recommendations appeared across multiple Books journeys, from Detail Pages and post purchase experiences to Kindle and broader discovery surfaces. But the experiences had evolved independently, with teams using different signals, terminology, and priorities. The challenge was not another carousel or component. It was creating a shared way to define, prioritize, and explain recommendations so future experiences could be more consistent and reusable.

Recommendation experiences felt connected to customers, but they were often designed and prioritized independently behind the scenes. One team might organize recommendations around author or series, while another emphasized genre, customer behavior, or a specific point in the journey. Each approach could solve a local problem well. Together, however, they created an ecosystem where the same word, recommendation, carried different meanings. Without a shared definition of similarity, decisions about placement, priority, and presentation remained subjective.

Rather than beginning with UI patterns, I stepped back and asked a more fundamental question: what actually makes two books similar? I reviewed existing recommendation experiences, worked through prior research with series customers, studied how other recommendation driven products communicated relevance, and held whiteboarding sessions with my product manager.

The synthesis revealed three insights: similarity is multidimensional, readers may connect through author, series, genre, characters, emotion, format, community, or topic. Customer intent changes the value of a recommendation, more of the same is different from the same feeling or help me explore further. And customers need context, a recommendation becomes more trustworthy when the relationship is explained.

Core similarities include the same author or series, similar authors or series, shared genre, setting, plot structure, style, or format. These are most useful when the customer is looking for continuity, for example on a Detail Page, immediately after purchase, or at the end of a series.

Adjacent similarities may include character dynamics, emotional tone, reviews, social conversation, community, or reader identity. They can support discovery when customers want something that resonates without being a near duplicate.

Contextual similarities can extend to related products or merchandise when the customer is exploring an interest rather than seeking another version of the same book.

Categorizing similarities was only useful if it changed how teams made decisions. I translated the taxonomy into a repeatable sequence: start with customer intent, identify the appropriate similarity type, consider the journey moment, select the placement and pattern, then determine what explanation the customer needs.

The framework helped teams think beyond what to recommend and consider when each type creates the most value: adjacent similarity early in discovery, core similarity near a title or purchase, contextual similarity after engagement.

Existing patterns could tell customers that they may also like a title without explaining why. Customers wanted to understand the signal behind a recommendation and how closely it related to the book they were viewing. Depending on the surface, that could mean similarity tags, supporting copy such as because you read, or lightweight metadata that names the author, series, genre, character, emotion, or topic connecting the titles.

A framework only creates value if other teams understand and use it. I documented the taxonomy, principles, and exploratory concepts in a strategy document, then used it to guide recurring conversations with Product, Engineering, Design, and partner teams that surfaced recommendations. I presented the work to Product Management leaders and partners, including the Detail Page and Kindle Unlimited teams, gathered feedback, and refined the thinking as their own roadmaps evolved. The document gave teams a shared object to critique and a vocabulary for discussing recommendation intent beyond individual UI patterns.

The strongest outcome was not a single recommendation component. It was a shared way to discuss what a recommendation was, which customer need it served, how strongly it related to the source title, and where it belonged in the journey.

The framework created three tangible outputs: a common vocabulary for Core, Adjacent, and Contextual similarity, a decision model connecting customer intent to priority, placement, and explanation, and a strategic foundation for future recommendation explorations across Books teams.

I would pilot the framework on a high traffic Detail Page surface, validate whether customers understand the similarity labels, and compare engagement and trust across recommendation types. I would also test where category boundaries blur, for example when a shared trope functions as a Core signal for one reader and an Adjacent signal for another.