Putting ChatGPT inside hundreds of millions of iPhones and getting almost nothing in return sounds impossible. But that’s exactly what happened. Court documents from OpenAI’s antitrust battle with Elon Musk’s SpaceXAI, reported by Android Headlines, reveal that OpenAI’s own internal teams described the Siri integration as “dramatically underperforming” within weeks of its December 2024 launch.
The failure came down to friction. Apple didn’t surface the ChatGPT option anywhere obvious. Instead, users had to find their way through a multi-step opt-in buried inside the Settings app. Most people who triggered Apple Intelligence features like Visual Intelligence, which uses the camera to analyze real-world objects, either dismissed the ChatGPT prompt or just opened the standalone ChatGPT app directly from their home screen. The path of least resistance won, and it didn’t involve Siri at all.
By January 2025, just a month after rollout, OpenAI had already cut its user growth forecast. The situation got worse from there. Strict privacy terms in the Apple agreement meant OpenAI couldn’t collect query data or use Siri traffic for model training. So there was no commercial upside from subscriptions and no technical upside from data. The deal delivered neither. By March 2026, conversations between the two companies had reportedly turned tense, with key sections redacted from the filings.
Apple also began building its Apple Foundation Models using Google’s Gemini Frontier models for distillation, which effectively sidelined OpenAI from the deeper infrastructure partnership it had hoped for. The two-year exclusivity window OpenAI had originally requested never materialized.
OpenAI made these numbers public for a specific reason: to destroy Musk’s legal argument before the case reaches its January 2027 trial date. Musk’s suit claimed that an illegal exclusive deal between Apple and OpenAI blocked his Grok chatbot from joining Siri, which supposedly prevented X from evolving into a “super app” capable of challenging the App Store and pressuring Apple to lower its 30% developer fee. Musk eventually dropped Apple from the suit entirely, leaving OpenAI as the only defendant.
To get the case dismissed ahead of trial, OpenAI brought in economic expert Dr. Catherine Tucker, who calculated that the share of AI consumers actually accessing ChatGPT through Apple Intelligence was effectively unmeasurable. The argument is simple and hard to counter: you can’t claim a monopoly harmed competition when the product at the center of that monopoly claim had almost no users.
OpenAI also demonstrated the deal was explicitly non-exclusive. Apple reserved the right to add other AI providers, and in practice never added rival extensions, but that’s Apple’s call, not evidence of a pact to exclude anyone. The core of Musk’s case rested on the assumption that the partnership was dominant and exclusionary. The internal performance data shows it was neither.
This situation matters beyond the lawsuit. The ChatGPT-Siri integration was supposed to be a blueprint for how AI companies could distribute their models through platform partnerships rather than fighting for direct consumer attention. It suggested that plugging into Apple’s install base was a shortcut to scale. The actual outcome suggests the opposite: if the on-ramp is clunky and the host platform hasn’t built real AI engagement habits, distribution alone doesn’t move the needle. Google’s Gemini faces similar questions as it gets embedded across Android and Workspace. Being present isn’t the same as being used.
