ChatGPT in Siri: Surprising Failure of Apple Intelligence

ChatGPT in Siri was meant to put OpenAI’s chatbot in front of every iPhone owner with a recent handset. OpenAI’s own lawyers now say it never came close. In a 59-page brief filed on 22 September 2026, the company told a federal court in Texas that Apple’s integration of ChatGPT was “dramatically underperforming” by the summer of 2025 and “persistently underperforming” by March 2026, and that its expert’s figures were “consistent with OpenAI’s internal view that Apple Intelligence saw minimal usage.”

The admission arrives in an unusual place. It is not a complaint against Apple, and it is not an earnings disclosure. It is OpenAI’s motion for summary judgment in the antitrust case that Elon Musk’s companies, X Corp. and SpaceXAI (formerly xAI), brought against Apple and OpenAI in August 2025. Musk’s side says the deal locked rival chatbots out of the iPhone. OpenAI’s answer is, in effect, that the deal was too small to lock anyone out of anything, because hardly anyone used it.

This article sets out what the filing actually says about ChatGPT in Siri, what it leaves under black redaction bars, why the integration struggled, why OpenAI is volunteering its own disappointment, how rivals grew while the partnership stalled, and what comes next for the lawsuit and for the extension inside Apple’s rebuilt Siri. Every date and figure below comes from the court record or from reporting cited in the references.

Table of contents

What OpenAI Told the Court About ChatGPT in Siri
Why ChatGPT in Siri Struggled to Draw Users
The Halo Effect OpenAI Expected From ChatGPT in Siri
Why OpenAI Is Arguing Its Own Deal Failed
What the Redactions Hide About ChatGPT in Siri
How Rivals Grew While ChatGPT in Siri Stalled
What Happens Next for ChatGPT in Siri and the Lawsuit
What the ChatGPT in Siri Story Means for AI Distribution Deals
Frequently Asked Questions About ChatGPT in Siri
References

What OpenAI Told the Court About ChatGPT in Siri

The key passages sit in the “Factual Background” of the brief, under two headings that tell the story on their own: “The integration of ChatGPT into Apple Intelligence struggles to draw users” and “The Apple integration fails to perform.”

The brief and the case it belongs to

The filing is OpenAI’s Brief in Support of its Motion for Summary Judgment, docket entry 403 in X Corp. and SpaceXAI LLC v. Apple Inc. et al., case 4:25-cv-00914-P in the US District Court for the Northern District of Texas, Fort Worth Division, before Judge Mark Pittman. It was filed on Tuesday 22 September 2026. Large parts of it are redacted, but the sentences describing how ChatGPT in Siri performed are readable, and they are unusually blunt for a company describing its own product.

“Off to a slow start” by January 2025

Apple switched on ChatGPT in Siri on 11 December 2024. By January 2025, the brief says, “Apple’s integration of ChatGPT appeared to be off to a slow start.” OpenAI reported internally that it lacked something about the integration’s outcomes (the missing word is redacted) and cut its forecast for the number of incremental logged-in weekly active users it expected to gain as a direct result of the deal. The new forecast figure is blacked out. So is the next sentence, which begins “By March 2025, OpenAI”.

“Dramatically underperforming” by summer 2025

The sharpest line comes in section F. “By the time Plaintiffs filed their Complaint, it was clear that Apple’s integration of ChatGPT was dramatically underperforming.” The complaint was filed on 25 August 2025. The brief then refers to something that happened in July 2025, quotes an OpenAI employee, and describes an action taken “on September 8, 2025”. All three are redacted. What survives is the verdict: within nine months of launch, OpenAI regarded ChatGPT in Siri as a failure against its own expectations.

“Persistently underperforming” by March 2026

The last dated entry is the most intriguing. “On March 25, 2026, with the Apple integration persistently underperforming, OpenAI told Apple that” — and the rest is redacted. The brief says OpenAI “reiterated” the point afterwards, also redacted. Seven weeks later, Bloomberg reported that OpenAI was weighing legal action against Apple over the partnership, which suggests the conversation was not a friendly one.

Date	What happened	Source
4 May 2024	Apple and OpenAI sign a non-exclusive Service Integration Agreement	OpenAI brief
10 June 2024	Apple announces Apple Intelligence and the ChatGPT option at WWDC	Apple
11 December 2024	ChatGPT in Siri goes live, switched off by default	OpenAI brief
January 2025	“Off to a slow start”; OpenAI cuts its user forecast	OpenAI brief
25 August 2025	X and xAI sue Apple and OpenAI; OpenAI says the deal is by then “dramatically underperforming”	OpenAI brief
12 January 2026	Apple and Google announce Gemini-based Apple Foundation Models	Google and Apple joint statement
25 March 2026	OpenAI tells Apple something (redacted) as the deal stays “persistently underperforming”	OpenAI brief
14 May 2026	Bloomberg reports OpenAI is weighing legal action against Apple	Bloomberg via MacRumors
14 September 2026	Musk’s companies drop Apple from the case with prejudice	Docket entry 385
22 September 2026	OpenAI files for summary judgment	Docket entries 402 and 403

Why ChatGPT in Siri Struggled to Draw Users

The brief does not offer a single cause, but it points at one immediately: the way ChatGPT in Siri was switched on.

ChatGPT in Siri was off by default, behind a multistep opt-in

“By default, the integration is turned off on iPhones,” the brief says. “To enable the integration, users must navigate a multistep process to affirmatively opt in, either through Siri or by locating the ‘settings’ page for the integration.” The sentence that follows, which starts “As a result”, is redacted, which is a strong hint that OpenAI’s evidence ties the opt-in directly to the numbers. Apple’s own support page confirms the route: open Settings, tap ChatGPT, tap Set Up, follow the on-screen instructions, then choose whether to sign in with a ChatGPT account.

ChatGPT in Siri was a handoff, not an assistant

Even once it was on, ChatGPT in Siri was an extension rather than a replacement. Siri answered what it could and offered to pass the rest to ChatGPT, asking permission before sending anything unless the user switched off “Confirm Requests”. Bloomberg’s May reporting added that users had to say “ChatGPT” to get results from OpenAI’s model, and that the answers shown inside Siri carried less information than the ChatGPT app itself.

The app was always one tap away

Anyone who already wanted ChatGPT had a better option on the same phone. The standalone app offered the full product, chat history and account features without a Siri hop in the middle. For those users, ChatGPT in Siri added a second, weaker doorway to something they could already open directly, which makes the low incremental numbers OpenAI describes easier to understand.

Apple Intelligence itself had a weak start

TechRadar’s reading of the filing is that the problem may not have been confined to ChatGPT in Siri at all, but to Apple Intelligence as a whole. Apple delayed the personalised Siri it demonstrated in June 2024, a delay that later produced the $250 million Siri AI settlement. ChatGPT also sat inside Visual Intelligence, the camera feature that uses computer vision to identify what the lens is pointed at, yet the brief’s own expert found that the share of consumers reaching ChatGPT through any of these entry points was small.

Step	Original Apple Intelligence Siri	Siri AI (beta, iOS 27)
Default state	Off	Off
Where to enable	Settings, then ChatGPT, then Set Up	Settings, then Set Up Extensions, then ChatGPT, then Use Extension
Account	Optional ChatGPT sign-in	Optional ChatGPT sign-in
Automatic handoff	Possible by turning off Confirm Requests	Not available; you must ask ChatGPT first
Siri suggesting ChatGPT	On unless Setup Prompts are turned off	Setup Prompts option not available
Minimum age	13, or the local minimum	13, or the local minimum

The Halo Effect OpenAI Expected From ChatGPT in Siri

OpenAI did not only count people who used ChatGPT in Siri. It also tried to measure what the Apple association did for the brand, and the brief is candid that neither number met expectations.

Direct effects versus indirect effects

From the outset, the brief says, OpenAI assessed the integration’s impact in two ways. Direct effects were use or subscriptions coming straight from Apple Intelligence users. Indirect effects were use or subscriptions through other channels that could be credited to “the intangible benefits of the partnership, such as promotion and brand association, sometimes referred to as a ‘halo effect.'” The analysis of indirect effects is described as showing they “were also” something; the word that follows is redacted.

The ChatGPT in Siri forecast OpenAI cut

The January 2025 ChatGPT in Siri forecast cut concerned “incremental logged-in weekly active users”, meaning new signed-in users OpenAI would not otherwise have gained. The original target and the revised one are both hidden. Elsewhere the brief says “OpenAI observed no meaningful benefit from such promotion”, a line that undercuts the idea that ChatGPT in Siri was quietly lifting ChatGPT’s numbers in the background.

What Bloomberg reported in May

The court filing’s account of ChatGPT in Siri fits what Bloomberg’s Mark Gurman published on 14 May 2026. OpenAI had believed the deal could generate billions of dollars a year in subscription revenue, which “hasn’t come close to happening,” and expected deeper integration across Apple’s apps and more prominent placement within Siri. “They basically said, ‘OpenAI needs to take a leap of faith and trust us,'” one OpenAI executive told Bloomberg. Another line was harsher: “We have done everything from a product perspective. They have not, and worse, they haven’t even made an honest effort.”

Apple’s version of events

Apple is no longer a party to the case, but its position in the record is consistent. It told OpenAI before WWDC 2024 that while it had no near-term plans to add other chatbots, it planned to announce that it would integrate others in future, and Craig Federighi said publicly that it was “looking forward to doing integrations with other models, including Google Gemini.” TechCrunch, reporting on the Bloomberg story, added that Apple had grievances of its own, including concerns about OpenAI’s privacy standards and irritation at OpenAI’s hardware push led by former Apple staff.

Why OpenAI Is Arguing Its Own Deal Failed

It is rare for a company to tell a court that one of its best-known partnerships flopped. OpenAI does it because, in this case, the flop is the defence.

Musk’s antitrust theory

On 11 August 2025 Musk posted on X that “Apple is behaving in a manner that makes it impossible for any AI company besides OpenAI to reach #1 in the App Store, which is an unequivocal antitrust violation,” and promised that “xAI will take immediate legal action.” The complaint followed on 25 August. It alleged that the Apple agreement was “exclusive”, that it “foreclosed” xAI from competing for customers, and that ChatGPT enjoyed “exclusive access” to “potentially billions of user prompts” that rivals needed as training data to improve their models.

The non-exclusivity clause

OpenAI’s first answer is the contract. According to the brief, OpenAI originally asked for a mutual two-year exclusivity period, and Apple refused. Apple’s term sheet struck the request and replaced it with “This agreement is non-exclusive by Apple. Apple may integrate other models.” The signed agreement, under a heading titled “Apple Discretion; Non-Exclusivity”, says OpenAI “acknowledge[d] and agree[d] that this Agreement is non-exclusive” and that Apple may integrate products with “the same or similar functionality” or build its own.

Foreclosure is the number that matters

The second answer is where the underperformance of ChatGPT in Siri becomes a legal weapon. Exclusive-dealing claims require proof that an agreement forecloses a substantial share of the market. OpenAI’s expert, Dr Catherine Tucker, calculated the share of generative AI consumers who reached ChatGPT through Apple Intelligence, using the plaintiffs’ own data and market definition, and found foreclosure shares that are redacted but described as “indisputably de minimis.” The brief notes that the plaintiffs’ economist, Dr David Evans, “does not dispute the math.”

Foreclosure benchmarks the brief cites, as a share of the relevant market

Usually required to prove unlawful exclusive dealing — 40% to 50%

Routinely held insufficient by courts — 10% to 15%

Called “quite insubstantial” by the Supreme Court in Tampa Electric — less than 1%

OpenAI’s expert figure for ChatGPT in Siri and Apple Intelligence — redacted, described as “de minimis”

“Where are the emails?”

The brief leans hard on what the plaintiffs did not produce. At a hearing on OpenAI’s motion to compel, Magistrate Judge Ray asked: “where are the emails? If this is such an awful monopolistic endeavor that’s costing the Plaintiffs billions of dollars, where are the emails from Mr. Musk?” OpenAI says that after months of discovery, none turned up. Its summary of the position is pointed: “Ironically, the only party that has produced evidence it is being disadvantaged by the Agreement is OpenAI.”

What the Redactions Hide About ChatGPT in Siri

The public version of the brief is a map with the most interesting streets painted over. Knowing where the gaps are matters as much as reading what is left.

The ChatGPT in Siri foreclosure share

The single most important figure, Dr Tucker’s foreclosure share, is blacked out in both the preliminary statement and the argument. The preliminary statement says the agreement covers “a tiny share” of the alleged market with the figure redacted, and a later sentence calls the number of new users “minuscule” before another redaction and the words “well below the 40% threshold”. The brief also quotes the Supreme Court’s line that foreclosure of “less than 1%” is “quite insubstantial” in a way that suggests the redacted figure is in that territory, but it does not say so in unredacted text.

The forecast figures

Neither the original user forecast for ChatGPT in Siri nor the reduced one is visible. That leaves no public way to say how far short the integration fell. It could have missed a modest target or a very large one, and the language (“slow start”, “dramatically underperforming”, “persistently underperforming”) describes the direction without the size.

The dated events

The July 2025 event, the 8 September 2025 action, and what OpenAI told Apple on 25 March 2026 are all redacted. So is a line in the argument that begins “By September 2025—less than a year after the integration went live—OpenAI told Apple” and then goes dark. Those passages probably hold the clearest account of how the relationship broke down, and they may surface later if the court unseals more of the record.

What the payment terms were

The brief confirms that the agreement has a section 5.1 dealing with what OpenAI received in return for the Apple integration, and redacts it. Earlier press reports said the arrangement involved no cash changing hands in either direction, with OpenAI trading technology for exposure. The court record neither confirms nor contradicts that.

Topic	Visible in the brief	Redacted
Usage	“Minimal usage”, “de minimis”, “slow start”	Every usage and share figure
Forecasts	OpenAI cut its incremental weekly user forecast	Original and revised numbers
Exclusivity	OpenAI asked for two years; Apple refused	Deposition quotes from Apple executives
Payment	A section 5.1 exists	What OpenAI got in return
Relationship	Dates: July 2025, 8 September 2025, 25 March 2026	What happened on each date
Model training	No training on logged-out prompts; consent needed for logged-in users	How many prompts were training-eligible

How Rivals Grew While ChatGPT in Siri Stalled

The second half of OpenAI’s argument is that the market kept moving without the iPhone slot. The brief gives several unredacted figures that make the point.

Gemini gained the most

Google had no chatbot integration in Apple Intelligence when ChatGPT in Siri launched, yet from the last quarter of 2024 to the second quarter of 2026, Gemini’s web traffic share rose from 5% to 27%, according to data from OpenAI’s expert cited in the brief. On 12 January 2026 Apple and Google went further, announcing that “the next generation of Apple Foundation Models will be based on Google’s Gemini models and cloud technology.”

xAI’s own valuation doubled

The brief uses the plaintiffs’ fundraising against them. xAI’s valuation went from $113 billion in July 2025, just before the lawsuit, to $230 billion in January 2026, when it raised $20 billion. It merged with SpaceX at a $250 billion valuation in February 2026, ahead of SpaceX’s June 2026 public listing. “Firms that are competitively stifled by an entrenched monopolist don’t double in value in six months,” the brief says.

Anthropic grew without Siri prompts

Anthropic is not integrated into Apple Intelligence either. Between March 2025 and May 2026, its reported valuation rose from $61.5 billion to $965 billion, “all without the Siri prompts that Plaintiffs say are so critical.” The brief also quotes Musk saying on 9 July 2026 that Anthropic is “obviously currently the leader in AI.”

ChatGPT Plus still costs $20

On pricing, the brief notes that ChatGPT Plus has cost $20 a month since it launched in February 2023, the same as the equivalent Google and Anthropic tiers and “about 33% below what xAI charges.” That is part of OpenAI’s argument that consumers have not been harmed, whatever happens at the top of the range, where OpenAI is preparing a $500 ChatGPT Pro Max tier.

Growth multiples from figures in the brief, while ChatGPT in Siri was live

Anthropic valuation, Mar 2025 to May 2026 — $965bn ÷ $61.5bn = 15.7x

Gemini web traffic share, Q4 2024 to Q2 2026 — 27% ÷ 5% = 5.4x

xAI value at the SpaceX merger vs July 2025 — $250bn ÷ $113bn = 2.2x

xAI valuation, July 2025 to January 2026 — $230bn ÷ $113bn = 2.0x

ChatGPT Plus monthly price, Feb 2023 to Sep 2026 — $20 ÷ $20 = 1.0x

What Happens Next for ChatGPT in Siri and the Lawsuit

The case has changed shape in the last fortnight, and so has the product at its centre.

Apple is out of the case

On Monday 14 September 2026, X Corp. and SpaceXAI told the court they had “resolved their claims” against Apple and moved to dismiss them with prejudice, which means they cannot be refiled. Apple did not oppose it, and the filing gave no reason and did not mention a settlement. Apple had asked the court to throw out the case in October 2025 and lost that motion, so its exit was negotiated rather than won. OpenAI is now the only defendant.

Summary judgment first, then a January 2027 trial

OpenAI asks the court to rule in its favour on every claim without a trial, on four independent grounds: no exclusivity, no substantial foreclosure, no antitrust standing, and no properly defined market. The plaintiffs will respond, and MacRumors reports that a trial is scheduled for January 2027 if the case survives. None of the arguments has been tested yet; the brief is OpenAI’s account, not a finding.

Siri AI changes the ChatGPT extension

Apple’s rebuilt assistant, announced on 8 June 2026 and now in beta with iOS 27, keeps ChatGPT as an optional extension but narrows its role. Apple’s support page says the automatic handoff option “isn’t available if you’re using Siri AI” and that “You must first ask ChatGPT before a prompt, photos, or files are sent to ChatGPT.” Siri no longer offers to fall back to ChatGPT on its own. Our guide to everything new in Siri AI covers the wider changes, which run on Apple Foundation Models developed with Google’s Gemini.

The two companies are fighting on other fronts too

OpenAI’s disappointment with ChatGPT in Siri is only one strand of a deteriorating relationship. Apple is separately suing OpenAI over alleged trade-secret theft by former Apple engineers, a case we covered in the six wildest claims in Apple’s lawsuit against OpenAI. And Bloomberg’s May report said OpenAI had hired an outside law firm and was considering at least a formal breach-of-contract notice.

What the ChatGPT in Siri Story Means for AI Distribution Deals

Beyond the courtroom, ChatGPT in Siri is one of the few hard data points on what a platform partnership is actually worth to an AI company. Businesses weighing their own integrations can take several lessons from it.

A default slot is not a distribution channel

The industry has treated placement inside an operating system as the ultimate prize, on the model of Google’s search deal with Apple. ChatGPT in Siri shows why that comparison breaks down. Search in Safari is on by default and used constantly. A chatbot extension that is off by default, needs a setup flow and hands off only when the host assistant decides to, is a much weaker thing, whatever the logo on the keynote slide.

Opt-in friction is a product decision

Apple built the opt-in design around privacy, and the support page spells out the protections: IP addresses are obscured, and requests from users who are not signed in cannot be stored or used for training. Every one of those protections also reduced how often ChatGPT was reached. Anyone negotiating an integration should treat defaults, prompts and confirmation steps as commercial terms, not interface details, because they decide the numbers.

Privacy terms limit the data prize

Musk’s complaint assumed ChatGPT was harvesting “billions” of valuable iPhone prompts. The brief says the agreement prohibits OpenAI from training on prompts from users who are not logged in, and requires explicit consent or notice before training on prompts from logged-in users, so “the number of training-eligible prompts that OpenAI receives is negligible to begin with.” Language models improve on data, but a privacy-first integration hands over very little of it.

Measure the partnership before and after you sign it

OpenAI tracked direct and indirect effects from the first month and revised its forecast within weeks. That discipline is what produced the evidence it is now using in court. Any organisation adding an AI assistant to its own product, or plugging into someone else’s, should define the metrics up front: incremental users, conversion, retention, and a baseline for the halo effect it hopes to see.

Frequently Asked Questions About ChatGPT in Siri

Is ChatGPT in Siri still available?

Yes. ChatGPT in Siri remains an optional extension in iOS 27, and it works with both the original Apple Intelligence Siri and the Siri AI beta. It is off by default. With Siri AI you enable it under Set Up Extensions, and you must ask ChatGPT explicitly before anything is sent.

Did OpenAI say how many people used ChatGPT in Siri?

Not publicly. Every usage figure, forecast and foreclosure share in the brief is redacted. The readable text describes usage as “minimal” and the foreclosure effect as “de minimis”, and says OpenAI cut its forecast for incremental logged-in weekly users in January 2025.

Why would OpenAI admit its own deal failed?

Because the admission helps it win the antitrust case. Musk’s companies must show that the agreement shut rivals out of a substantial share of the market. If very few people ever used ChatGPT in Siri, the agreement cannot have shut anyone out of much.

Does this prove Apple Intelligence failed?

It shows that OpenAI’s internal view was that Apple Intelligence saw minimal usage, at least as a route to ChatGPT. It does not include Apple’s own usage data for features such as Writing Tools or notification summaries, and Apple is no longer a party to the case.

Is Apple still a defendant?

No. The plaintiffs dismissed their claims against Apple with prejudice on 14 September 2026 after saying they had resolved them. OpenAI is the remaining defendant.