Product A waitlist. Active conversations with brand managers across skincare, supplements, B2B SaaS — segments where LLM-recommended purchase intent is fastest-growing.

Product B paid-pilot conversations. Two enterprise brands in active scoping for performance-priced 6-week pilots, post-funding.

Sample audit available. Anonymized illustrative audit (rank #4 → #1 in six weeks) demonstrating the convergence loop. Full audit shared per-vertical under NDA.

Product A — SaaS. $200/seat/month entry. Mid-market brand teams average ~10 seats → ~$24K/yr/logo. Enterprise tier (custom, $5–20K/mo) for 20+ users + SLA.

Product B — performance-priced. $X per LLM-attributed click, $Y per qualified lead. Pricing tied to Product A simulator-predicted lift. Target gross margin 50–70%.

Expansion. Land at Product A self-serve seats; expand to enterprise; cross-sell into Product B once Product A demonstrates predicted lift. One stack means one deployment touches both revenue lines.

Surrogate model — the 97% claim. Fine-tuned on observed ChatGPT-served rankings; held-out evaluation across N intent categories yields 97% Spearman rank-correlation. Architecture, training procedure, scatter plots & CIs under NDA. Re-evaluated quarterly; degrades on very-low-popularity queries.

Adversarial foundations. Zou et al. (2023) — arXiv:2307.15043. Universal & transferable adversarial attacks on aligned LMs (GCG). The discipline behind the adversarial-string lever.

GEO foundations. Aggarwal et al., GEO: Generative Engine Optimization, ACM SIGKDD 2024. The peer-reviewed founding paper of the GEO field.

Brand-safe agent guarantees. Granziol et al., ICML 2026 — safety–efficacy trade-off in autonomous LLM agents (addressing Anthropic open problems). Underpins the brand-safety claims on the Two Products slide.

Surrogate-model architecture. Bayesian + random-matrix-theory-grounded design choices document why the model generalizes faithfully across query distributions. Diagram + training procedure under NDA.