Podcasts on Google Assistant
Personalized endorsements using activity-based signals
Google Assistant endorsements are sometimes initially written using machine learning. The copy isn’t always relevant, actionable, or consistent.
Challenge
Copy didn’t entice the user to listen to the suggested podcast and the user didn’t understand why a podcast was suggested. Some endorsements were offensive or inappropriate, based on an incorrect understanding of the category type.
Solution
Provide the user with appropriate endorsements based on their listening habits and other activity-based signals to increase product engagement.
How I helped:
Wrote new copy
Edited existing copy to align with new style guidelines
Used personalization based signals to tailor endorsements based on user preferences. This meant for some strings I was writing 4 versions of each endorsement for 4 different user segments.
Made blocklist decisions for several category types that were offensive
Personalized endorsements for broad categories of podcasts
Original copy
Comedy podcast
Suggested copy
No personalization: If you like comedy podcasts
Inferred mid-personalization: You may like comedy podcasts
Inferred high-personalization: You like comedy podcasts
Activity-based personalization: You’ve listened to other comedy podcasts
Why
While “Comedy podcast” explains the content type, it doesn’t encourage the user to listen to the suggestion.
I used personalization based signals to tailor endorsements for users based on their preferences.
Simple endorsement based on past activity
Original copy
If you’re interested in Nerdificent
Suggested copy
For fans of Nerdificent
Why
The show endorsement is based on previous listening history signals so the word “if” isn’t strong enough to explain why the recommendation is relevant.
I chose “fans” to make the recommendation more personal and warm.