An algorithmically generated group of items, often shown as follow-up recommendations or related actions
E-Commerce・Generative Design・Social Media・Content Distribution

Work In Progress

Our Elements guide is still in progress, and therefore lacks full visual and technical assets. We hope to release them by summer of 2020. Thanks for reading Lingua Franca!


An assortment is a collection of items that are presented to a user, usually in the form of content that ‘you may also like’ or information that users ‘might be interested in’. Typically, this prompt is given in the hypothetical tense, to encourage users to engage if they choose. Content provided in an assortment element is typically personalized to the user’s preferences or context, in order to further increase the user’s engagement with the system. However, assortments run the risk of manipulating users and promoting negative habits and behaviors. Users should be able to dismiss or re-label individual items to improve future recommendations.


The assortment is one of the earliest uses of machine learning in digital interfaces. They allowed users to discover more content while creating additional page impressions and keeping users in their ecosystem. Early examples include Amazon and Yahoo, two pioneers of web-based AI interaction design.

One of the reasons for the popularity of the assortment as a technique is its ability to condense complex information into an interlinked web of similar content. It allows information to be algorithmically grouped in the same way that products are grouped on a grocery store shelf to make them easier to find and browse.


Like any recommender system, the algorithm used to generate assortments can create different impacts on users (see Intuition). Depending on the context and training data, an assortment may seem rather unhelpful and generic, or it could seem serendipitous and exciting (see Design Tradeoffs).

The ‘Harry Potter Problem’

This phenomenon of generic recommendations is often called the ‘Harry Potter Problem’, since it is akin to recommending Harry Potter to every customer in a bookstore. This phenomenon can emerge for a variety of reasons, such as low ‘salience’ that encourages safe recommendations, a lack of detailed data on user preferences, or from the ‘cold start’ problem where new users tend to receive generic recommendations.

Different kinds of content may require separate algorithms or datasets, given that browsing habits differ between content categories. An algorithm recommending similar books may benefit from genre tags, while an algorithm powering e-commerce recommendations may benefit from better shopper demographics.

A crucial consideration with assortments is the level of salience in the recommendations (often termed entropy for technical reasons). This refers to the level of boldness, or surprise in the recommendations. This factor is not always reducible to a single number to tweak (though, surprisingly, it sometimes is). Think about this factor, and how it may enable certain kinds of user journeys that relate to task completion, content discovery, and recommendation quality.