Siri Recap vs a recording you keep

Apple states the boundaries on its own support page

Apple announced Audio Intelligence on 9 September 2026. Its support page calls it “four new features”: Sound Recognition, Music Recognition with Shazam, Live Rewind and Siri Recap. Two of them touch speech, and the support page is precise about what each one does with it.

Every row above was read on Apple’s Audio Intelligence support page on 20 September 2026. Two of the four features were still described as arriving in beta then, so wording, schedules and language coverage can move before they reach your wrist. Apple gives no date beyond “later this year” and no usage numbers.

Live Rewind catches 15 seconds, then lets go

Double-press the Digital Crown and the watch shows a text snippet of what was just said. Fifteen seconds. Apple writes that the feature “only activates when you deliberately double-press the Digital Crown” and “doesn’t continue to transcribe in the background”.

• A chime plays from the watch speaker, even on silent and even with headphones connected. Apple says this alerts people nearby.
• A full-display animation and a microphone indicator appear at the same moment.
• The snippet disappears from the watch roughly 30 seconds after the display dims, unless you save it to the Siri app or tap “Ask Siri”.
• Speech recognition runs on your Apple Watch and iPhone. Saving a snippet or asking Siri sends it to Private Cloud Compute.

That covers one sentence you missed in a loud room: a platform number, a dosage, a surname. It gives you no route back to the start of the conversation. Ten minutes ago is out of reach.

The recap is a summary, and it expires in 7 days

Apple describes the feature as something that “ambiently takes notes on your conversations throughout your day”. The result is short. Apple compares it to “notes you might jot down after a conversation”.

A transcript does get made on the way, and you never get to read it. Sound enters a protected buffer in the S11 chip’s Secure Exclave on the watch and travels encrypted to the Secure Exclave on your paired iPhone, where Apple writes that it is “decrypted, transcribed, and condensed”. Only the condensed text leaves for Private Cloud Compute, “designed to be less than half the length of the original transcript” in Apple’s words, and the summary is what comes back to you.

Open the Watch app on iPhone and turn Siri Recap on. Apple says you can then set a schedule by time, by location, or both, in its own examples “only at work” or “never at night”, or leave it manual and switch it yourself. Control Center turns it off wherever you are.

A summary you do nothing with is gone in a week. Save or export one and it stays in the Siri app, where deleting it removes it from your synced devices. Apple says only you can reach a saved summary. Recaps also omit sensitive material such as financial data and government-assigned identifiers, so a number you wanted may not be there.

Match the situation to the right one of the three

Two run by themselves. The third you have to start. Read down the left column for the thing you are actually trying to do.

An older Apple Watch won’t get Siri Recap, but it can still record

Apple lists the hardware on the same support page. Live Rewind and Siri Recap need Apple Watch Series 12 or Ultra 4, paired with a supported iPhone, and they need Apple Intelligence and the Siri AI beta. Apple’s iPhone list starts at iPhone 16 and leaves out iPhone 16e. Both features start in English, and neither is available to anyone under 13. Apple also says Siri AI will not be available at first in the EU on iOS, iPadOS and watchOS.

Phrase’s watch app needs watchOS 26 or later. Apple’s watchOS 26 guide lists Series 6 through Series 11, SE (2nd generation) and SE 3, and Ultra, Ultra 2 and Ultra 3, and the newest watches run later versions. A recording hands off to Phrase on iPhone and becomes a note with the audio attached. You start and stop each one yourself. Nothing listens through the day on a schedule the way Siri Recap does.

Apple’s feature requirements were read on 20 September 2026 and its watchOS 26 model list on 25 September 2026. Both lists can change when the beta ships.

Start the recording yourself when the words have to survive

The gap is not a flaw in what Apple shipped. A memory aid and a record are different jobs, and Apple built the first one on purpose. The moment you need to reread a sentence, show who said it, or listen again after the meeting has faded, you need a file that exists.

1. Decide before the conversation starts. Nothing reaches backwards once it is over.
2. Tell the people in the room. Consent rules differ by country and by state, and one sentence at the start costs less than finding out later.
3. Record where the sound is. Voice Memos on Apple Watch saves to the watch and syncs to iPhone, where those recordings sit under a separate Watch Recordings entry.
4. Name it the same day, with the date, who was there, and the subject. A folder of files called New Recording is its own problem.

Phrase records from iPhone or from its Apple Watch app, and the audio stays attached to the note that comes out of it. The transcript is segmented with timestamps and optional speaker labels, tapping a segment jumps playback to that second when the audio is on the device, and the note stays in your library until you delete it.

A failed task leaves no half-written note behind, and media-to-note processing runs in the cloud, so the note appears once the task finishes. Keep the original audio either way. It is the only thing that settles an argument about what was said.

If the watch is where the conversation happens, set up recording from the Phrase app on Apple Watch so the audio and the note arrive together instead of a summary that expires.

Questions readers are asking

Does Siri Recap make a transcript? Not one you ever see. Apple writes that Siri Recap “doesn’t produce a transcript” and “generates a brief, high-level summary” instead. Its privacy section also says the audio is “decrypted, transcribed, and condensed inside the iPhone Secure Exclave”. So words are transcribed, inside a compartment where, in Apple’s words, data “can’t be accessed by the operating system, apps, you, or even Apple”. What reaches the Siri app is the summary. For the sentence someone actually said, you need a recording made for that purpose.

Is Siri Recap private? Apple’s stated design keeps no audio: “Audio is not recorded or stored”, and summarising runs on Private Cloud Compute. Summaries auto-delete after 7 days unless saved. Protection for people nearby rests on no stored audio and no speaker attribution.

What is Live Rewind on Apple Watch? A 15-second text snippet of what was just said, shown when you double-press the Digital Crown. A chime and a full-display animation signal that it ran. The snippet disappears about 30 seconds after the display dims unless you save it to the Siri app.

What is Audio Intelligence on Apple Watch? Apple’s name for four features running on the S11 chip: Sound Recognition, Music Recognition with Shazam, Live Rewind and the daily recap. Sound Recognition switches on from iPhone Settings, Accessibility, Sound & Name Recognition, Sound Recognition, then “Detect using Apple Watch”.

Does it work on iPhone? Apple documents Audio Intelligence as an Apple Watch capability with the paired iPhone doing part of the work: the watch sends encrypted audio to the Secure Exclave on the phone, which transcribes and condenses it. You switch both features on from the Watch app on iPhone.