What Was Grammarly Thinking?

Grammarly launched an 'Expert Review' AI feature that framed editing advice as inspired by named authors and academics, then deactivated the feature within a day after backlash and a reported class-action suit by journalist Julia Angwin; CEO Shishir Mehrotra apologized and said the company will strongly defend against the claims. The incident raises reputational and legal/IP risk for Grammarly/Superhuman Platform and underscores execution and product-liability issues for generative-AI features; likely limited to reputational/legal costs rather than market-moving financial exposure.

Analysis

This episode is a useful canary for an emergent economic wedge: provenance and consent become monetizable safety features for generative-AI products. Firms that can credibly prove training provenance, auditability, and an opt-in/compensated dataset supply chain will capture a premium from enterprises and publishers over the next 6–24 months; expect procurement cycles to lengthen but deal sizes to increase for vendors that offer those controls.

A wave of IP and right-of-publicity litigation is the most immediate tail risk and will impose nonlinear costs on smaller, data-hungry model providers. Defense and remediation (retraining, content removal, licensing) are multi-million-dollar line items that can compress margins by several hundred basis points within 3–12 months, and that dynamic will push investors toward vertically integrated platforms with captive datasets.

Second-order winners are human-in-the-loop marketplaces and quality-control software: buyers who want legal-safe output will pay for human editors or provenance metadata, supporting revenue growth at platforms that match skilled workers to content tasks. Conversely, thin-margin consumer writing apps and pure-play LLM vendors face two-way pressure—customer churn from reputational hits and higher unit costs to buy/verify licensed training data.

Catalysts to watch that will re-rate the group: (1) major licensing deals between platform and publisher (0–9 months), (2) class-action outcomes or regulatory guidance from the FTC/DOJ (3–18 months), and (3) adoption metrics for provenance features in enterprise RFPs (6–24 months). A favorable licensing market or clear regulation would rapidly reverse the flight-to-safety trade into a content-licensing boom.

Key Decisions for Investors

• Long MSFT (12–24 months): buy shares or 2027–2028 calls. Rationale: Microsoft’s enterprise distribution, Azure compliance tooling, and integrated editor stack position it to capture premium enterprise spend on provenance-enabled AI. Risk/reward: target +15–25% upside if enterprise adoption accelerates; downside -15% if regulation materially limits model capabilities.
• Long ADBE (6–18 months): buy shares or a call spread into next major earnings. Rationale: Adobe can productize content authenticity and paid creative datasets, converting brand trust into pricing power. Risk/reward: expect +20% upside on feature monetization; policy/regulatory reversals or slower enterprise uptake could stall gains.
• Long UPWK (3–12 months): buy shares. Rationale: human-editor marketplaces benefit as firms pay for legal-safe, human-reviewed content and forensics services; short-term demand spike as customers backfill AI shortcomings. Risk/reward: +25–40% upside if adoption rises; automation risk could halve upside over 18+ months.
• Short C3.ai (CAI) via 6–9 month put spread: Rationale: pure-play model vendors are most exposed to margin hits from litigation and dataset licensing costs; a 30–50% downside is plausible if class actions proliferate. Risk/reward: structured put spread to cap premium outlay while keeping >2:1 potential payoff vs cost.