Choose and test a Bluetooth microphone for voice AI

End active sessions and disconnect the Bluetooth accessory. In the assistant, use a quiet, short phrase such as “Route test: seven yellow folders.” Verify that it appears correctly as input. Then send a text request for a short spoken response, with spoken output enabled if your installed version supports it.

Record the two outcomes separately: capture succeeded or failed, playback succeeded or failed. If the built-in route fails, solve that first. Otherwise you may blame an accessory for a permission or network issue that exists without it.

Pair or reconnect the accessory using the system controls, then confirm ordinary playback reaches it. Apple documents selecting a Bluetooth playback destination, but that destination alone is not evidence of which microphone a third-party app has selected.

Start a fresh assistant session after the connection is stable. Avoid moving the accessory out of range or opening its charging case midway through the test. Changes during capture can introduce a separate interruption problem, making the result harder to interpret.

Repeat the same phrase at a normal voice level. Compare transcript clarity and response timing with the baseline. If you are testing microphone location, keep the phone and headset positions fixed and use ordinary speech; do not rely on extreme shouting or rubbing microphones.

Use only an input selector actually offered by the installed app or operating system. If no selector is available, record that limitation rather than claiming you forced a specific source. A headphone symbol generally describes a route, not a verified microphone measurement.

If audio plays in the headset but the transcript is empty, investigate input selection, permission or audio-session handling. If the transcript is correct but replies play from the phone, investigate output routing. If both stop when another app opens audio, note the competing activity and repeat without it.

Try a second known-working accessory only after the first comparison is documented. If one headset fails and another succeeds under otherwise identical conditions, accessory compatibility is worth investigating. That still does not identify a hardware defect without further tests.

For an important session, choose the route you have actually tested. Keep a text fallback ready, and end the conversation before reconnecting accessories. Headphones can also reduce the chance that the assistant hears its own speaker output.

When reporting an issue, include accessory model, device/OS/app version, connection order and the separate capture/playback results. GPT Realtime Assistant’s public mobile sessions require internet, so include whether text requests still worked. This avoids confusing network failure with a Bluetooth microphone problem.