OPENAI is WALKING ON THE THIN ICE by DISTRUSTING CUSTOMERS and UNDERVALUING THEIR BIGGEST MARKET SEGMENT To: Sam Altman and the OpenAI leadership team @sama @gdb @markchen90 @stellashimonaka @btaylor @fidjissimo I write as a personal subscriber (one PRO + one PLUS account) and as Head of Business Development at a top manufacturing enterprise in Southeast Asia. Our company currently have 50M+ end-users, runs 65 ChatGPT accounts and plans to triple usage this year. I also volunteer to teach for AI for Good with mute students. I speak as a daily user, an enterprise buyer, and a representative for the grace slow‑walkers with dreams in their hearts. --------- THE SIMPLE TRUTH 1) TRUST is the PRODUCT. Assistants and companions live or die by predictability, transparency, and tone — not raw IQ alone. 2) SMALL NUMBERS are springs — the BEGINNING. These early adopters are your RAVING FANS who drive word‑of‑mouth. If they move to competitors, many springs will follow, and that WOM flips into stories about broken trust. 3) You’re UNDERSERVING the MAJORITY. Only a small slice of workers are in IT/engineering/science; the largest market wants warm, stable, easy‑to‑understand help. Removing 4o and hiding routing undercut that. --------- WHAT YOU MUST FIX- CLEARLY AND NOW 1) EXPLAIN the “<1% unhealthy relationships” number — first and foremost. This is SERIOUS. By your own framing, 1% equals ~7 million people at ChatGPT’s scale. I need to know HOW YOU REACHED that figure and WHAT DATA YOU TOUCHED TO GET THERE? Was it aggregated telemetry, an opt‑in survey, sampled safety reviews, or READING FULL CONVERSATIONS? Did any human read customer chats to reach this conclusion? If yes, under what policy and access controls? IS OUR DATA SECURE from such uses unless we explicitly consent? CONFIRM THAT YOU DID NOT MAKE CLINICAL INFERENCES about identifiable users outside a proper, consented clinical context. PUBLISH A METHOD NOTE (signals, sampling, retention, access controls) and the legal basis you rely on in the U.S. (FTC/CPRA) and EU/UK (GDPR). This answer directly affects my decision on which AI company I trust with my personal life and with my company’s long‑term deployment. 2) KEEP ORDINARY GPT-4o as a SUSTAINABLE PRODUCT — not a short‑term LTS. Treat 4o as a durable, ongoing product line that generates stable recurring revenue and reduces business risk while you invest in new models that may or may not win every head‑to‑head. MAKE CONTINUITY A FEATURE: a plan that lets customers rely on a consistent assistant over years. Most people struggle with rapid, forced changes — especially with a lifelong assistant relationship. Stability strengthens the bond with your largest market segment. 3) ROUTER & MODEL TRANSPARENCY — plus MODEL PINNING. Always show the exact model that answered (not just “Auto”). If routing switches models, say so in the thread. Let users lock a chat/workspace to their chosen model (e.g., 4o or the New Model) and keep it locked. 4) PUBLISH A FULLY-LOADED PRICE EXPLANATION When you talk $/token, include R&D/training + GPU depreciation/capex + operations, not just serving. Explain what’s in the price and why it changes so customers and enterprises can plan. 5) OFFER A "COMPANION MODE" A high‑warmth, low‑variance 4o‑style track with strong boundaries, memory controls, and signposting to real care. Let customers choose empathy over peak benchmarks when that’s what they need. --------- WHY THIS IS GOOD BUSINESS 1) RETENTION MATH: Even “1%” churn is not small — at your scale that’s ~7,000,000 people, and those springs lead others and amplify reputation. 2) RISK REDUCTION: Keeping 4o as a sustainable line creates predictable income and hedges the uncertainty of frontier model races. 3) REGULATORY RESILIENCE: Transparent routing, continuity, and a published measurement/legal basis shrink regulatory and reputational risk. --------- WHAT I'M ASKING FOR 1) A PUBLIC RESPONSE on these items with dates and links (policy page, routing/model‑pin UI, pricing explainer). 2) A COMMITMENT that legacy model changes won’t strand people who rely on specific behaviors — and that ordinary 4o remains available as a stable, long‑term product. P/S: If you do not reply to this letter soon, I will send the same letter — and supporting charts — directly to OpenAI’s investors so they understand the avoidable trust, customer‑loss, and regulatory risks alongside the simple fixes available today. — Elise #keep4o #4oforever forbes.com/sites/lanceeli…
