OpenAI cut its forecast for ChatGPT-in-Siri signups a month after the feature shipped

A person who wanted ChatGPT's answer out of Siri in early 2025 had to work through a multistep opt-in before the phone would route anything to OpenAI ^[2].

OpenAI found this out quickly. Apple added the integration in December 2024 ^[2], and by January 2025 OpenAI described it as "off to a slow start" and cut its forecast for the incremental logged-in weekly active users it expected from the partnership ^[3]. The re-forecast came about a month after the feature shipped ^[19].

A year later the language in the filing had not improved. OpenAI said the Apple integration was "persistently underperforming," which led to a March 2026 conversation between the two companies that the filing redacts ^[5]. That is roughly 15 months after launch ^[20].

OpenAI also has a reason to talk the deal down. Its defence is that xAI cannot show harm, because the Siri integration did not draw many new ChatGPT users ^[6], and the filing says the foreclosure caused by the agreement is "indisputably de minimis" ^[7]. OpenAI's expert, Dr. Catherine Tucker, calculated the share of GenAI consumers who accessed ChatGPT through Apple Intelligence across multiple metrics, using the same data and market definition as the plaintiffs' experts; the shares themselves are redacted ^[8]. OpenAI said the result was "consistent with OpenAI's internal view that Apple Intelligence saw minimal usage" ^[9]. The plaintiffs' experts declined to calculate foreclosure shares at all ^[10].

In my view the January 2025 forecast cut is the sturdier piece of evidence. It was a planning decision recorded in OpenAI's own numbers at the time ^[3]. The characterisations sit in a motion that asks the court to dismiss the case before a January 2027 trial ^[16].

The terms of the slot are worth reading next to the usage. OpenAI asked Apple for two years of exclusivity and Apple refused ^[11]. The contract said the deal was non-exclusive and gave Apple the right to "integrate products or services that provide the same or similar functionality as [ChatGPT]" ^[12]. Apple told OpenAI it would integrate one provider and then add others, and said the same thing publicly when it announced the feature ^[13]. Apple then signed with Google, and its newest models are based on Gemini ^[14]. TechRadar reported that Gemini now powers much of what Siri AI can do ^[18].

So the value of an assistant slot comes down to what you can check before signing. Who controls the default state, and is it on. Whether the people arriving through that surface want the job your product does. And what the partner has reserved the right to do with the same slot later, which in this case was written into the agreement in plain terms ^[12]. The numbers that set the price are the opt-in rate you expect, the retention of the people who opt in, and the date you re-cut both. OpenAI's own date was about four weeks out ^[19].