New York Times personalized article recommendations: product and engine design
Company: Google
Role: Product Manager
Category: Product Design & Strategy
Difficulty: medium
Interview Round: Other
You are a product manager working on The New York Times. Design a content recommendation engine that improves which articles and other news content readers are shown as recommendations.
Explain whom the engine serves, where its recommendations appear, what it should optimize for, and how you would measure whether it works.
```hint Decide what a good recommendation means for news
A click is a weak signal for a news reader. Think about which reader behaviors show that a recommendation really served them, and which decisions editors would not hand over to an algorithm.
```
### Clarifying Questions
- Which surfaces are in scope: the home page and app feed, the end of each article, newsletters, push alerts, or all of them?
- Which readers matter most: paying subscribers, registered readers who have not subscribed, or anonymous visitors arriving from search and social media?
- Is the main goal more reading per visit, more return visits, keeping subscribers, or turning readers into subscribers?
- How much of the front page must stay chosen by editors, and are any kinds of stories off limits for personalization?
### What a Strong Answer Covers
- A goal tied to the publisher's subscription business and its editorial mission, with a chosen reader segment and the surfaces where recommendations will appear.
- Reader pain points with today's story discovery, and solutions prioritized into a first version.
- How the engine works at a product level: candidate sources, the ranking objective, editorial and diversity rules, freshness, and cold start for new or anonymous readers.
- Success metrics that go beyond clicks, with guardrails for diversity, editorial priorities and trust, plus an experiment plan.
- Risks such as filter bubbles, drift toward sensational stories and privacy, and how each is controlled.
### Follow-up Questions
- Editors worry the engine will bury important but less popular stories. How would you balance editorial picks against personalization?
- Your test shows clicks up, subscriber cancellations unchanged and more complaints about repetitive stories. How do you read the result, and what do you change?
- How should recommendations work for a first-time visitor who lands on one article from a search result with no history at all?
Overview: A product design prompt asking a product manager to design a news recommendation engine for The New York Times. It tests choosing readers, surfaces and goals, defining what the engine should optimize beyond clicks, balancing personalization with editorial judgment and diversity, handling cold start, and measuring success with guardrails.