OpenAI’s ChatGPT integration with Siri did not become the user-acquisition shortcut either company may have hoped for. In a court filing connected to the ongoing dispute brought by Elon Musk’s companies, OpenAI described Apple’s ChatGPT integration as “persistently underperforming,” after initially calling its start slow and reducing its forecast for added logged-in weekly active users.

The filing matters beyond the question of how many people chose to use ChatGPT through Siri. It pushes back on a central allegation in the lawsuit: that Apple and OpenAI struck an exclusive arrangement that harmed competing generative-AI services, including xAI. OpenAI’s position is that its contract with Apple was expressly non-exclusive, that Apple retained the right to add similar providers, and that the integration’s limited uptake makes the claimed competitive damage difficult to establish.

Much of the underlying material remains redacted, so there is no public numerical accounting of usage, forecasts, or the internal discussions mentioned in the filing. Still, the disclosed language provides an unusually blunt view of a high-profile AI distribution partnership that, at least by OpenAI’s own account in this legal context, failed to produce meaningful growth.

What OpenAI says happened after ChatGPT came to Siri

Apple added ChatGPT support to Siri in December 2024. The setup required users to complete a multistep opt-in process. That detail is central to OpenAI’s explanation for the disappointing results: every additional action between a user and a feature can reduce participation.

By January 2025, OpenAI said the rollout was off to a slow start. It lowered its expectations for the number of incremental logged-in weekly active users expected from the partnership. In plain terms, that metric is not simply the number of people who might encounter ChatGPT through Siri. It concerns users who newly sign in and then use the service during a week—an important distinction when measuring whether a platform integration actually expands an AI company’s direct user base.

OpenAI later said it had become clear that the Apple integration was dramatically underperforming by the time Musk’s companies filed their antitrust action. A separate part of the filing describes the performance as persistently weak and refers to a March 2026 conversation between Apple and OpenAI. The substance of that conversation is not public because it was redacted.

That leaves important questions unanswered. The disclosed record does not provide a precise explanation of whether the low results chiefly reflected the opt-in design, lack of user awareness, infrequent use of the relevant Siri requests, users choosing not to connect accounts, or some combination of those factors. Nor does it provide an adoption number against which to judge the scale of the shortfall.

What the filing does establish is OpenAI’s argument that this was a small channel rather than a dominant gatekeeper to generative AI. That framing is strategically important because the same low adoption OpenAI regards as commercially disappointing also supports its defense against claims of market foreclosure.

OpenAI says it asked Apple for two years of exclusivity, but Apple declined. The agreement instead stated that it was non-exclusive and allowed Apple to integrate products or services with the same or similar functionality as ChatGPT.

The distinction is more than contract wording. Exclusivity generally means one party agrees not to offer a competing service or supplier for a defined function or period. A non-exclusive arrangement allows the platform to work with other providers. OpenAI says Apple told it that the company intended to begin with one provider and add more over time, a position Apple also expressed publicly when it introduced the feature.

Apple has since signed a deal with Google, and its newest models are based on Gemini. Those facts reinforce OpenAI’s claim that Apple was not locked into ChatGPT as its sole AI option. They do not, by themselves, settle every antitrust question raised in the case, but they directly challenge the allegation that Apple granted OpenAI a protected and exclusive Siri position.

For users, a non-exclusive model potentially means more choices in the future, though the filing does not spell out how Apple will present, select, or prioritize providers. It also does not establish that every provider will receive identical technical access or visibility. “Non-exclusive” is a meaningful contractual fact, but it is not a full blueprint for how a multi-provider AI experience will operate in practice.

Why low usage is part of OpenAI’s defense

The lawsuit alleged that Apple’s arrangement with OpenAI was exclusive and damaged xAI’s growth and customer acquisition. OpenAI disputes both ideas. Its argument is straightforward: the agreement did not exclude rivals, and even under an assumption favorable to the plaintiffs—that Apple users choosing ChatGPT through Apple Intelligence could be considered unavailable to competitors—the actual number involved was too small to matter competitively.

“Even if the Court assumes that Apple users who elect to use ChatGPT through Apple Intelligence are foreclosed from OpenAI’s rivals (which they are not), the amount of foreclosure caused by the Agreement is indisputably de minimis.”

Foreclosure is an antitrust term for conduct that allegedly limits rivals’ access to customers, markets, or distribution. In this dispute, the theory would be that an Apple-integrated ChatGPT option denied competing AI firms an opportunity to reach Apple users. OpenAI rejects the premise and then argues, alternatively, that the alleged effect was de minimis—too small to be legally meaningful.

The filing says OpenAI’s expert, Dr. Catherine Tucker, calculated shares of generative-AI consumers accessing ChatGPT through Apple Intelligence using the same data and market definition relied on by the plaintiffs’ experts. The resulting shares are redacted, but OpenAI says they were minimal across the measured approaches. It also says the plaintiffs’ experts did not calculate foreclosure shares.

Because the figures are hidden, readers should be careful not to turn OpenAI’s characterization into a verified public usage total. The filing is an advocacy document: it is presenting facts and analysis in support of OpenAI’s request that the court dismiss xAI’s claims before trial. Yet the basic logic is easy to understand. A claim that a partnership diverted a consequential portion of a market becomes harder to prove if the feature at issue did not attract many users.

The opt-in problem is a practical lesson for AI integrations

The Siri arrangement illustrates a recurring challenge for AI features embedded inside larger products: availability is not the same as adoption. A tool may be technically integrated into a widely used ecosystem, but users still need to understand what it does, decide that it is useful for a particular request, and be comfortable completing any required consent or account steps.

Here, users had to opt in through multiple steps. That can be valuable from a privacy and consent perspective, since it makes the user’s choice explicit. But it can also create friction. In product design, friction means an obstacle—however modest—that interrupts someone on the way to completing an action. A prompt, confirmation screen, account sign-in, or handoff to another service can each reduce the number of people who continue.

The filing does not say that opt-in was the sole cause of poor usage, and it would be a mistake to treat it as proven. Still, OpenAI specifically identified the multistep process in describing the slow start. The broader implication is that AI companies seeking growth through device partnerships cannot rely solely on placement. They must make the value of the handoff clear enough for users to take the extra step.

That challenge is relevant as AI assistants increasingly move into voice-driven hardware and ambient computing. The product question is not merely whether an assistant is present; it is whether people know when to invoke it, trust it with their requests, and can do so without a cumbersome transition. That same tension is visible in the wider push toward AI-enabled wearables, including plans for an AI agent on smart glasses, where a voice interface can reduce steps but still has to earn a place in daily routines.

What the filing does and does not resolve

Apple is no longer a defendant in this case after Musk’s companies dropped their claims against it earlier this month. OpenAI remains the sole defendant. It is asking the court to dismiss xAI’s claims before the trial currently planned for January 2027.

The filing does not determine the outcome. It offers OpenAI’s factual and legal response, and many potentially illuminating details are redacted. The court will assess the competing claims, the contracts, expert analysis, and whatever evidence is available through the case.

Still, the document puts two connected realities into focus. First, Apple’s ChatGPT arrangement was designed as non-exclusive, with room for other AI providers and services. Second, OpenAI says the integration itself drew minimal use and did not create the hoped-for influx of logged-in weekly users. Those points can coexist: a deal may be highly visible, strategically significant, and yet deliver little direct user growth.

For the AI industry, that is a useful corrective to the assumption that a major platform tie-in automatically guarantees adoption. Distribution can be powerful, but consent flows, account requirements, user habits, and the clarity of the product experience can matter just as much. In this case, OpenAI is now relying on that disappointing adoption story not as a product triumph, but as a key reason the alleged anticompetitive harm was negligible.