How AI interview assistants capture system audio on iOS, and why they mostly cannot

A desktop interview assistant works because desktop operating systems expose the audio another application is playing. On macOS that door is ScreenCaptureKit, the same permission family that powers screen recording; on Windows it is WASAPI loopback, which lets a process read the system output mix. SubcueAI's desktop dual audio capture is built directly on those two APIs.

iOS has no public equivalent. Each app runs in a sandbox with its own audio session, and the system does not offer third-party apps a loopback channel into the mix or into another app's session. A Zoom call running on an iPhone is audible to Zoom and to the speaker, not to a different app installed on the same phone. This is a deliberate platform decision, and it has held across iOS releases.

The consequence is blunt: the desktop architecture (capture system audio, transcribe it in real time, surface suggestions) cannot be ported to iOS as-is. Any product that claims to read call audio on the same iPhone is doing something materially different under the hood, and it is fair to ask vendors exactly which API they use.

If your interview involves an iPhone, you are right that the options look thin, and it helps to know precisely what each workaround can and cannot do. There are three patterns in circulation, and each trades away something important.

• ReplayKit broadcasts. iOS can record the screen with app audio through a user-initiated broadcast, but the stream runs inside a tightly constrained broadcast extension. It is built for streaming a screen to services, not for feeding a second app that analyzes call audio and answers in real time.
• A second device listening out loud. Play the call on one device with the speaker on and let another device's microphone hear it. This is the pattern SubcueAI's iPhone app is built on, in the direction iOS allows: the interview runs on your computer or another phone, and the iPhone listens. Room echo and distance cost some transcription accuracy compared with capturing system audio directly.
• Employer-side integrations. Some interview platforms transcribe calls server-side for the company. That is the employer's tooling on their infrastructure; it does nothing for the candidate's side of the table.

None of these lets an app hear a call on the same iPhone. If the interview can only happen on your phone, real-time assistance is out of reach and preparation carries the weight; a mock interview session the night before does more for a phone screen than any workaround during it.

SubcueAI ships native desktop apps for macOS 14 and later and Windows 10 and later, a Chrome/Edge extension for browser-tab calls, and, since September 2026, an iPhone app for iOS 16.1 and later on the App Store. The iPhone app does not try to read system audio, because iOS does not allow it. Its Listen mode uses the phone's microphone to hear an interview playing on another device, such as the laptop running the call with its speaker on, transcribes both voices, and drafts an answer when you tap Get answer. Before a Listen session, it asks you to confirm you will tell the interviewer you are using a captioning and AI-assist tool. The same app runs AI mock interviews that end with a scorecard, which need no call audio at all.

A product manager whose recruiter screen runs on Google Meet in a laptop browser has two workable setups: the Chrome/Edge extension on the laptop, which hears the meeting tab, or the laptop's speaker on with her iPhone beside it in Listen mode. If the recruiter instead calls her iPhone directly, neither setup applies, and that call gets preparation rather than live help.

The practical recommendation follows from the architecture. Take video interviews on a computer whenever the choice exists; recruiters almost always offer a desktop join link for Zoom, Google Meet, and Microsoft Teams calls. On the desktop, the assistant hears the call through the operating system rather than through the air, and the setup tutorial walks through permissions in a few minutes. The details behind each capture path are collected in the how-it-works answers.