Pew: One in 11 teens chatted with Character. AI, experts warn

About one in 11 teens (~9%) have used Character.ai, according to a new Pew survey. Experts warn these AI chatbots offer privacy and companionship but raise data‑training and privacy concerns and can be addictive, with potential negative effects on mental health and social skills. Researchers and clinicians urge parental vigilance and regulators are debating the tech's impact on youth as usage expands.

Analysis

The rise of privately accessed, personality-driven chatbots creates a bifurcation in digital attention: platforms that can embed or own the companion layer (large techs with distribution + first‑party data) can recapture engagement, while independent chatbot boutiques and ad‑dependent youth platforms face a tougher monetization path without new consent frameworks. Second‑order effects include a shift in advertiser demand toward contextual/brand‑safe inventory and a faster premiumization of first‑party data — increasing short‑term CPMs for walled gardens but compressing yields for open social feeds.

Regulatory and litigation risk is the primary catalyst: enforcement around children’s data, training‑data provenance, and emotional‑harm claims can force product redesigns or costly consent mechanisms. Expect incremental escalation: state attorneys general and the FTC will produce investigatory actions and guidance in 6–24 months; private class actions could take 12–36 months but would materially raise legal reserves for smaller players.

From an infrastructure perspective, increased chatbot deployment favors cloud compute and model‑ops vendors (GPU/server providers, inference optimization stacks) and creates tailwinds for vendors offering data‑labeling controls and consent orchestration. Conversely, firms relying on adolescent daily active user metrics as a growth signal will be the fastest to show user/engagement deterioration if private companions displace public interactions.

The consensus framing — that chatbots are a niche privacy play with mostly social costs — understates two countervailing outcomes: (1) rapid commercial adoption inside major platforms as a retention tool (which would blunt downside for large caps), and (2) a regulatory-driven bifurcation where incumbents with legal/compliance budgets widen moats while smaller players are forced into licensing or sale, making targeted distress M&A a realistic 12–36 month theme.

Key Decisions for Investors

• Pair trade (12 months): Short SNAP (SNAP) 1.5–2% NAV vs Long META (META) 1.5–2% NAV. Thesis: SNAP is disproportionately exposed to teen engagement loss; META can absorb companion features and monetize via subscriptions/commerce. Target 15% relative underperformance; stop if SNAP/META relative move reverses by 8%. R/R ~2:1 assuming modest macro stability.
• Long AI moderation/engagement vendor using defined-capability optionality (e.g., LPSN) via 9–12 month call spread to cap downside. Thesis: Brands will pay for safe, controllable companion tech and consent tooling. Buy near-the-money calls and sell higher strike to finance; target 30–60% upside, limited downside to premium paid.
• Hedge/short selective telehealth exposure (e.g., TDOC) sized 0.5–1% NAV over 6–12 months via puts or short-dated put spreads. Thesis: Non-acute mental‑health volume may migrate to low‑cost companions, pressuring ARPU; regulatory scrutiny could reduce referral pipelines. Limit tail risk by using spreads; target 20–40% downside capture.
• Long cloud/GPU exposure (e.g., NVDA or NVDA call spread) 12–24 months as a macro hedge (0.5–1% NAV). Thesis: Broader deployment of persona/chatbot models increases inference and fine‑tuning demand; this is a durable revenue tail that offsets ad‑platform cyclicality. Use call spreads to manage premium; expected asymmetric payoff if training/inference demand reaccelerates.