How Tonic is Experimenting With Reader Vibes

Remember the last time you opened an app and felt seen, attacked, or plainly delighted by something that was recommended to you, seemingly by a machine. That perfect recommendation can be a rewarding feeling, but it also makes you wonder, "What about me made it think I wanted this?"

Ever since we set out to build a private personalization stack that gives you transparency & control of your data and recommendations, we’ve been experimenting with ways to make the experience of a personalized reading app more, well, personal. We want to explain what is being understood about your activity, in a way that’s fun and makes you feel seen.

For the past several months, we’ve been experimenting internally with an explainability feature we’re calling Vibes. Vibes – also, a mood or atmosphere – are what Tonic thinks you’re giving off based on your reading history, likes and dislikes.

What kind of Vibes can Tonic detect about you? Well, you can be a combination of nosy, fiery, competitive, meditative, playful and many more. This idea was inspired by a core role-playing video game mechanic, where your characters can develop into archetypes, bringing with them specialized skills and capabilities. Our Editorial Director, Bassey Etim, developed the initial set of 11 “personality” archetypes in May and they’ve stuck since.

In our first pass, Vibes were mapped directly to our content clusters – groups of inherently similar content in our recommendable catalog. We soon realized that wouldn’t work. Any Vibe could cut across multiple topics. For example, you can be in a Meditative mood, and enjoy pieces that are thought-provoking rather than on a certain subject. Similarly, a single story can appeal to multiple Vibes for different reasons.

This led us to a whole new tagging & machine learning project. Our engineering and editorial folks came together to build new infrastructure to allow curators to tag individual pieces of content and, after countless hours back-tagging, training several classifiers to make predictions.

As we continue this experiment, we’re excited about Vibes breaking new ground for personalized reading experiences in two ways: one, they scratch an itch we have to quantify & know ourselves – we’re no strangers to lookback moments, like “Spotify Wrapped”, and taking personality quizzes on the internet– and two, they’re a powerful way to exert control of our recommendations. The latter is something we really want to tackle next.

We’re all aware that personalized feeds offer you no transparency or recourse and tend to push you into filter bubbles. They all overlook that we’re multifaceted and our tastes, and even reactions to a piece of content, change over time. Yet, most reading apps use topical & interest labels and shoehorn you into those topics the more you engage. Static labels and filter bubbles means a wide variety of content that we’re actually in the mood to read will not come into our consumption.

Because Vibes have an inherent temporality, we can offer readers insight into their reading choices, and give them a way to change what they read to suit themselves without forcing subject matter. You can jump from a Fiery vibe to a Blissful one, and get a recommendation fitting that mood.

In our first all-hands meeting of 2020, we shared our most recent vibes with one another, just to see if they felt right. Graham, our Head of Comms, was a Blissful Leader (so true); Dan, our Head of Engineering, was a Nosy Planner (of course he was!), and myself? Turns out I was being Playfully Nosy. Thanks Tonic, I feel seen.

If you’re interested in helping us test out Vibes, send an email to

published by:
Laima Tazmin
January 31, 2020