Skip to content

Product1 publisher3 min readPublished

TechCrunch's iOS 27 hands-on credits the working Siri to Google's Gemini models

A hands-on review says the assistant Apple teased over two years ago now handles multi-step requests and on-screen questions, with Google's Gemini named as the model underneath. The evidence so far is one person's account.

The Product Desk · Product desk

Illustration accompanying TechCrunch's iOS 27 hands-on credits the working Siri to Google's Gemini models

What happened

  • Apple is shipping the advanced Siri it teased alongside Apple Intelligence more than two years ago, after a run of delays.
  • The reviewer reports Siri taking multi-step instructions, including finding hidden iOS 27 features and writing them into a note, and drafting messages from context in files, messages and email.
  • Asked in the Camera app what drink the reviewer was holding, Siri answered green juice first and named Jaljeera or Aam Panna only after being told the restaurant was Indian.

Compiled by The Product DeskSomething wrong?How this is made

Why it matters

  • exposure If the review is right, Apple's most visible consumer AI feature sits on top of a model from the company it competes with in phones and search, so a change in Google's pricing or policy reaches iPhone owners through Siri.
  • constraint Reliability depends on how people named their own notes and files, so a rollout needs a data-hygiene pass before it needs a training session, and the support queue will show up as 'it put it in the wrong place'.
  • decision Anyone weighing a Siri integration this quarter is working from one reviewer's account, with the accuracy unmeasured. That makes an early rollout your own pilot, and it should be budgeted as one.

The workaround for the one thing that kept breaking was to rename a note. TechCrunch's reviewer wanted Siri to file coffee roasters found on Reddit into a specific note, had repeated trouble getting the right one, and settled it by simplifying the note's name [8]. Note names are chosen for humans. Anyone planning to tell a team "just ask Siri" on Monday should expect that step to surface first.

The Gemini claim rests on one sentence in that review, which says Siri is "Built on the back of Google's Gemini models" [2]. The same passage says it "is much smarter than its previous version that was built on extended knowledge and a shaky ChatGPT integration" [3]. TechCrunch did not say which requests are answered on the device and which go to Google, and the review includes no accuracy figures [15].

Apple teased this version when it introduced Apple Intelligence over two years ago, then shipped it after multiple delays [1]. The review ran on September 14, 2026 [17], which places that tease in September 2024 or earlier [18].

The usage evidence is one person on a main device since the public beta, plus a second device on the developer beta before it [4]. The review does show depth: Siri pulling a delivery conversation out of messages to draft a returns note [12], answering questions about what is on screen, including the oldest player in a World Cup line-up and a startup's funding and competitors from its website [13], and taking a multi-step instruction to find hidden iOS 27 features and write them into a note [6]. iOS 27 also gives Siri its own app with previous chats and a Settings pane for voice and expressiveness [14].

The review is specific about where it missed. Sports answers had hiccups in the earlier developer builds [5], the note targeting needed the rename [8], and in the Camera app Siri said "This is green juice" [9] before the reviewer specified an Indian restaurant, after which it offered Jaljeera or Aam Panna and was right [10]. Three corrections across the interactions described [16].

The handoff is the untested part. Apple has positioned the new Siri to work with third-party apps as developers update them, and the review treats the smoothness of that handoff as an open question [11].

Which tasks are worth handing over? Does a wrong answer show up immediately, and does success depend on Siri resolving a name inside the user's own data? The drink in the viewfinder is the cheap case, wrong on the first pass and corrected in one follow-up [9][10]. Filing to a named note is the expensive case, because the failure goes unseen and depends on labels the user chose years ago [8]. Drafting a message and sending it [7] fails after the message has gone to someone else. A pilot belongs in the first case, and the third stays out of the rollout until Siri's name resolution is documented.

What to watch

  • Whether Apple confirms the Gemini arrangement or publishes which Siri requests are answered on device.
  • Whether the Siri-to-app handoff holds up once developers ship apps updated for it.
  • Whether note and file targeting gets more reliable without users renaming their own content.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories