Character.AI PM interview: safety vs. engagement as product strategy

Character.AI’s PM interview is not a standard AI product round. The company is currently navigating wrongful death lawsuits, state AG enforcement actions, a founder departure, a valuation that fell from $2.5B to roughly $1B, and a forced pivot away from its most-used feature for its largest audience segment. Every product sense question is implicitly a test of whether you understand that tension and can reason through it without flinching. Candidates who show up with a generic social product playbook (optimize DAU, add notifications, A/B test onboarding) will be identified quickly.

The good news: the process is fast. Average time to hire is 11 days, difficulty is rated 2.75 out of 5 (easiest of all roles at the company), and 87.5% of candidates report a positive experience. This is a lean startup-speed process, not a FAANG-style gauntlet.

The numbers you need to know cold

Before the interview, these figures should be second nature:

• 20M MAU and 75 minutes average daily engagement per user. ChatGPT users average around 8 minutes per session. The engagement gap is real and extraordinary.
• 463 million site visits per year, making it the second-most-visited AI tool after ChatGPT. Average session time: 17 minutes 23 seconds.
• 18M+ user-created bots and a core user base that is 75.56% ages 18-34 (Gen Z).
• $9.99/month c.ai+ subscription. Estimated 2025 revenue: $50M. That gives an ARPU of roughly $0.72 against those engagement numbers. The product is deeply used and poorly monetized.
• Minor chat removal: November 2025, Character.AI removed open-ended chat for under-18 users entirely. February 2026: confirmed global chat shutdown for teens, with minors redirected to interactive story and video creation tools. 78% of transitioning minor users adopted the alternative features.
• Catalyst: Sewell Setzer III, 14, died by suicide in February 2025 after months of interactions with a Character.AI chatbot. The wrongful death lawsuit and subsequent Kentucky AG enforcement action (January 2026) drove the safety overhaul.
• Competitive set: OpenAI GPT Store, Meta’s in-app AI characters, Replika ($24-30M revenue, 2M MAU), Chai Research (6M MAU).
• Founding situation: Noam Shazeer and Daniel De Freitas returned to Google in 2024. The company now runs on open-source models (Meta, DeepSeek) rather than proprietary ones. As of mid-2025, the company is evaluating a sale or additional fundraising.

Drop these in product sense answers. Vague references to “Character.AI’s large user base” without specifics read as surface prep.

The rounds

Recruiter screen. Standard background and fit check. The recruiter is confirming you know the product and can articulate the business model. “High engagement, thin monetization, safety pivot” is the one-sentence thesis you should be able to deliver without prompting.

Hiring manager or PM interview. This is where the real test begins. Expect a product sense question structured around Character.AI’s own product: how would you improve it, how would you measure success, what would you cut. The question is a vehicle for surfacing how you reason about the safety-engagement trade-off, not a generic product design exercise.

Cross-functional or panel round. May include engineering or policy stakeholders depending on the role. Behavioral questions here lean on ambiguity, competing priorities, and decisions made under legal or regulatory pressure.

Case or take-home. Some roles include a written case or live strategy question. Expect something that forces you to grapple with monetization given the current ARPU and subscription model, or with the minor user pivot and what it means for product direction going forward.

The distinctive signal: holding the tension

Character.AI’s core product challenge is not technically hard in 2026. Feasibility is not the constraint. Building an AI companion that generates deep, emotionally intimate relationships at scale is clearly possible; the company has already done it. The question is whether you can build it in a way that is viable (subscriptions grow, the company survives lawsuits and regulatory scrutiny) and genuinely lovable in the 2026 sense: meeting emotionally isolated Gen Z users where they are, without crossing into dependency, harm, or manipulation.

The PM who treats this as a standard engagement optimization problem, or who mentions “adding safety guardrails” as an afterthought feature, fails. The PM who understands that the company’s most engaging feature for its largest audience was also the feature that drove a wrongful death lawsuit and the first state AG enforcement action has the right frame. The safety overhaul is not a constraint on the product. It is the product strategy.

Questions that have been asked or are likely

• “How would you define success for Character.AI’s subscription product given the current ARPU?”
• “How would you improve Character.AI for your most valuable user segment?”
• “Minor chat was removed for teens. What would you build next for the 18-34 core base?”
• “How do you think about the trade-off between engagement depth and dependency risk in a social AI product?”
• “Character.AI runs on open-source models now. How does that change your product and competitive strategy?”
• “A feature drives a 20% increase in session length. When should you ship it, and when should you not?”

What clears the bar

Name the tension explicitly and show you’ve thought through the full stack. Know the numbers (20M MAU, 75-min engagement, $0.72 ARPU, $50M revenue, minor chat removal, 78% alternative feature adoption). Frame your north star metric around paying user behavior rather than raw engagement. Treat the safety pivot not as damage control but as a product strategy decision that defines what Character.AI is for in 2026. Have a concrete view on what the product should be for the 18-34 core base now that minor users are out of chat entirely. That is the whitespace the company is actually trying to fill.

For the viable/lovable frame on consumer social AI, see lovable, not just usable. For the broader case on proving out viability in a thin-ARPU consumer AI business, see proving viability. For how the consumer-vs-enterprise PM axis applies to a role like this, see consumer vs. enterprise PM.