OpenAI's filing traces weak ChatGPT adoption on iPhone to an integration Apple shipped switched off

An integration that ships switched off needs the user to find out the feature exists and then go turn it on. The Financial Times reports that OpenAI's filing points to Apple's implementation as one possible reason adoption was slow, because the integration was turned off by default ^[5].

A bundle deck counts platform placement as reach. Placement behind an off switch reaches only the users who open settings.

The integration launched in 2024 ^[1], and the "dramatically underperforming" conclusion arrived in the summer of 2025 ^[4]. The latest that launch could have been is December, and the earliest summer month is June, so at least six months of usage sat between OpenAI's first forecast cut and that conclusion ^[14].

Multiple conversations in the filing are redacted ^[6], and the report does not put user or subscriber numbers behind the word underperforming ^[15]. OpenAI is making the argument in a brief whose purpose is to show the Apple partnership foreclosed nobody. Apple and OpenAI have both argued that the deal was not exclusive as the complaint alleged ^[8], and OpenAI told the court that SpaceX's IPO registration statement was "replete with disclosures diametrically opposed" to the competitive harm the suit claims, Reuters reported ^[9]. A defendant in that posture has an obvious use for evidence that its own distribution deal did very little.

For anyone negotiating placement inside another company's operating system, two variables set the value: how deep the placement goes, and what state the feature ships in. OpenAI got the depth and an off switch ^[5]. Depth is worth whatever an opt-in setting converts, so a partner who cannot get default-on written into the agreement should forecast the deal as opt-in traffic, with a slow first month as the base case. That discipline costs you a smaller number in the plan and a harder conversation with whoever wrote the original one.

Craig Federighi said in 2024 that Apple ultimately wanted users to be able to choose between different AI models, and pointed to Google Gemini as a possible future integration ^[12]. In the unredacted portions of the filing, OpenAI executives were concerned that Apple could "want to move faster" to secure another AI partner ^[6]. Apple has since turned to Google's Gemini models to help power the new Siri AI ^[11].