Transcription suddenly got worse or feels less accurate

Available on: Mac, Windows, iOS, and Android.

If dictation suddenly loses words, produces garbled text, or feels less accurate, compare a short test recording with its transcript. That separates an audio problem from a language or transcription problem.

Quick checks

Restart Flow: quit and reopen it, then dictate a sentence longer than three seconds — shorter desktop recordings can return empty text, and taps under about half a second count as accidental.
Check the active microphone: a newly connected headset, webcam, or Bluetooth device can change the input between desktop dictations.
Record in another app — Voice Memos on Mac/iPhone, Sound Recorder on Windows — and listen for low volume, distortion, or missing first and last words. If the recording is unclear outside Flow too, fix the microphone or system input first; if the audio is clear but Flow's text is wrong, continue to playback and language checks.

Check desktop audio

Check	Mac	Windows
Microphone permission	System Settings → Privacy & Security → Microphone	Settings → Privacy & security → Microphone; allow desktop app access
Input device and level	System Settings → Sound → Input	Flow's sound-settings shortcut opens the classic Sound control panel → Recording.

Enable Flow's microphone access using the permission settings above, then reopen Flow.
Select your intended microphone in the system sound settings.
Set the input level so the meter moves clearly without constantly reaching maximum. Flow does not adjust the system input level, room noise, or microphone distance for you.
Open Flow's Settings → General → Microphone → Change.
Select the intended mic and speak; its live level meter should move.
Test the same phrase after each change. If the text is accurate, stop; otherwise compare a built-in or wired mic against Bluetooth and continue below.

Note: Bluetooth microphones can delay audio capture and clip opening words — on Mac, let the mic become ready before speaking. With a MacBook lid closed, select an external microphone; the built-in mic is unavailable, and its no-audio error includes an action that opens the microphone picker.

If Flow switches microphones

In the Microphone dialog, Auto-detect uses the system default, and Edit ranking lets you drag microphones into priority order. Selecting a mic places it above connected peers.

Flow can switch when a mic disconnects or a higher-priority mic connects; a new higher-priority mic takes effect on the next dictation, not mid-recording. Disconnecting the mic in use stops dictation with Mic disconnected, whose action inserts the last text. Once a ranking exists, a new mic outside that list can prompt Switch or Don't Switch after dictation ends; the prompt excludes AirPods and virtual devices.

Virtual or loopback inputs can capture silence or another source, so Flow uses them only when you select them explicitly.

If the microphone test fails

During dictation, silence warnings appear after about five seconds and escalate after about fifteen seconds with Select microphone and Contact us; they clear when audio returns and do not appear during microphone testing.

Permission errors offer access settings; for a missing microphone, select the notification's button to open Sound settings instead. An unavailable device offers Choose microphone, an unreadable device offers Troubleshoot, and a mic that takes longer than eight seconds to open offers Restart Flow.

For clipped beginnings, see Missing first words in transcriptions.

Listen to the original and retry

Desktop recordings are in Home → Recent activity. Use Search past transcripts to locate an entry, then hover over or keyboard-select it to reveal its actions.

Important: Desktop audio expires after 14 days while transcript text remains. Save or report a recording before its retention window ends.

Select Play audio to compare the recording with the text. If the audio has the same gaps, check mic position, Bluetooth, input volume, and background noise.
Choose More options (⋯) → Retry transcript if the audio is complete and clear. ⋯ → Extract audio saves a WAV file.
Compare the updated text; if it remains wrong, check languages below.

When retry is unavailable or fails

All retries need retained audio: after 14 days, playback is hidden and extraction is disabled, and missing audio prevents playback, download, and retry. Dismissed recordings shorter than about half a second are never saved.

The menu's Retry transcript has no minimum duration but is disabled while the row loads or dictation is active. Inline Retry on empty or failed entries needs at least five seconds of audio and an entry newer than 14 days; Recover on dismissed or still-processing entries only requires that age limit.

Retry replaces the entry's text, but can leave the original unchanged if it is interrupted or taken over by a new dictation. For “Retry failed. Please try again.”, retry after dictation stops.

If text exists but never reached the app, follow text-pasting troubleshooting instead.

Check phone audio

iPhone

If Flow shows “We can't hear you”, Go to Settings opens microphone settings; Flow rechecks permission when you return.

Reopen Flow and confirm microphone permission in iOS Settings.
Disconnect Bluetooth to compare against the built-in mic, keep the phone's microphones uncovered, and test in a quieter space.
Swipe a History entry to use Play when playable audio remains, then Retry and compare the text with the audio.

History also offers Report, but no audio download. If audio was never recorded, it cannot be recovered — dictate again.

Android

Confirm your intended microphone or hearing aid is connected before recording. If accuracy drops partway through, the input may have switched: stop, confirm the active input, and start again.

Tap Go to Settings on “We can't hear you” to allow microphone access.
Test another phrase and compare the result.
Choose ⋮ → Retry Transcript on a failed entry when captured audio exists, then check the updated text.

Without captured audio, Retry Transcript is hidden. “Retry failed. Audio file not found.” or “Retry failed. Audio file missing.” means the recording is unavailable; for “Retry failed. Please try again.”, retry.

Sessions stop at five minutes with a one-minute warning, which can explain a cut-off passage.

Check language and repeated substitutions

Open desktop Settings → General → Languages; iOS Settings → Set language; or Android Settings → Languages.
Select the languages you speak, then dictate again to check the result.

On desktop, keep at least one language or Auto-detect (99 languages) selected; Auto-detect applies when no individual languages are selected, and switching it off in the same dialog session restores your previous choices. If you dictate in a language that doesn't match your selection, Flow suggests checking your language settings at most once a day and offers a link that opens them; Don't show again stops the suggestion permanently.

Tip: If you speak Ukrainian and your speech comes back as Russian, select both Ukrainian and Russian. Flow then double-checks speech that sounds Russian before choosing the final text, which makes Ukrainian far more accurate; genuine Russian still comes back as Russian. This extra check can add a slight delay.

Regional variants use picker labels such as Hinglish, English - British, Chinese - Simplified (简体中文), and German - Swiss (Deutsch), and selecting one deselects its conflicting variant; see language setup.

If names or technical terms are repeatedly wrong, review your dictionary words and corrections. On desktop, Add to vocabulary → Correct a misspelling creates a rule that automatically replaces the incorrect spelling.

If a result still looks unpolished, retry when available or divide long thoughts into shorter recordings. Desktop dictations stop at 20 minutes with a warning at 19 minutes, and very long recordings can lose final words after you stop; see accuracy and known limitations.

Still stuck?

Contact support if clear, complete audio repeatedly produces incorrect text after these checks, or if retry and feedback submission fail.

Open Help → Talk to support on desktop, which also opens the Help Center.
Select Report an issue and Request Type → Issue.
Describe your microphone, languages, affected recording, expected text, result, and checks completed.
Choose Send message.

The form automatically includes platform, time-zone, and app details, plus your preferences file when within its size limit. If sending fails, Flow copies your message to the clipboard so you can paste it and try again.

Send transcription feedback without requesting a reply

Entry feedback improves transcription; it is not a support request and does not receive a reply.

Select the entry's flag icon, Send feedback, on desktop, or Report on an Android home-screen entry.
Describe what you expected instead; an empty report cannot be sent.
Choose Send feedback and check for “Report sent!” or a failure message.

Desktop reports automatically include any retained recording, transcript versions, selected languages, and app, timing, and duration details. Recordings of any length are sent.