TikTok Cold Start: Using AI to Personalize a Brand-New Viewer's First Feed

Quick Overview

A TikTok product sense question on using AI to improve recommendations for brand-new viewers who have no watch history yet. It tests reasoning about cold-start signals, exploration versus exploitation, onboarding friction, retention-focused metrics, and safety and privacy guardrails for new and younger users.

TikTok Cold Start: Using AI to Personalize a Brand-New Viewer's First Feed

Company: TikTok

Role: Product Manager

Category: Product Design & Strategy

Difficulty: medium

Interview Round: Other

You are a product manager at TikTok. A person who opens the app for the first time has no viewing history, so the For You feed has very little to personalize with during those first sessions. How would you apply AI to make the recommendations these newcomers see better from the start? Explain which newcomers you would focus on, what you would change in the product, and how you would know the first experience improved. ```hint Inventory what you already know A newcomer is not a complete blank. List what the app can observe before the first video plays and during the first few swipes, then ask which product moments could gather more of it without slowing the person down. ``` ### Clarifying Questions - Do "new users" include only first-time installs, or also logged-out viewers and returning people on a new device? - Which outcome matters most: enjoyment in the first session, retention over the following weeks, or turning logged-out viewers into account holders? - Should newcomers under 18 be handled separately, given the extra content and safety restrictions that apply to younger users? - Are there limits on the data the feature may use, for example contacts or information from other apps? ### What a Strong Answer Covers - A chosen newcomer segment and a clear explanation of why cold start matters so much for a feed-first product like TikTok - The signals available before any history exists, and product moments that collect more of them without adding friction - Specific AI techniques (such as planned exploration, video content understanding and learning from similar users) translated into changes a newcomer would notice, and prioritized - Metrics that cover the first session and longer-term retention, with guardrails on safety, feed diversity and creator exposure - Risks around privacy, stereotyping from demographic proxies, unsuitable content for minors, and a feed that narrows too quickly ### Follow-up Questions - How would you decide how much of a newcomer's first feed should be exploration rather than videos you already expect them to like? - If an onboarding interest picker raised first-session watch time but lowered 30-day retention, how would you explain it and what would you do? - How would your approach change for a logged-out viewer whose activity you cannot tie to an account?

Overview: A TikTok product sense question on using AI to improve recommendations for brand-new viewers who have no watch history yet. It tests reasoning about cold-start signals, exploration versus exploitation, onboarding friction, retention-focused metrics, and safety and privacy guardrails for new and younger users.

TikTok logo
TikTok
Oct 1, 2026
mediumProduct ManagerOtherProduct Design & Strategy
0
0

You are a product manager at TikTok. A person who opens the app for the first time has no viewing history, so the For You feed has very little to personalize with during those first sessions. How would you apply AI to make the recommendations these newcomers see better from the start?

Explain which newcomers you would focus on, what you would change in the product, and how you would know the first experience improved.

Clarifying Questions Guidance

  • Do "new users" include only first-time installs, or also logged-out viewers and returning people on a new device?
  • Which outcome matters most: enjoyment in the first session, retention over the following weeks, or turning logged-out viewers into account holders?
  • Should newcomers under 18 be handled separately, given the extra content and safety restrictions that apply to younger users?
  • Are there limits on the data the feature may use, for example contacts or information from other apps?

What a Strong Answer Covers Guidance

  • A chosen newcomer segment and a clear explanation of why cold start matters so much for a feed-first product like TikTok
  • The signals available before any history exists, and product moments that collect more of them without adding friction
  • Specific AI techniques (such as planned exploration, video content understanding and learning from similar users) translated into changes a newcomer would notice, and prioritized
  • Metrics that cover the first session and longer-term retention, with guardrails on safety, feed diversity and creator exposure
  • Risks around privacy, stereotyping from demographic proxies, unsuitable content for minors, and a feed that narrows too quickly

Follow-up Questions Guidance

  • How would you decide how much of a newcomer's first feed should be exploration rather than videos you already expect them to like?
  • If an onboarding interest picker raised first-session watch time but lowered 30-day retention, how would you explain it and what would you do?
  • How would your approach change for a logged-out viewer whose activity you cannot tie to an account?