Reviewed Jun 2026 · some cells re-verified Sep 2026 · policies are checked weekly for changes · stated — every cell is read from Wispr Flow's published policy; none has been tested (run by us) yet.
1Frictionstated
The privacy toggles are self-serve (Dictation Cloud Storage, the training-share toggle, Context Awareness — all under Settings → Data and Privacy), and account deletion is self-serve on desktop and iOS — but Android deletion runs through a support ticket, rights requests are email-only with no stated windows, and neither GPC nor Do Not Track is mentioned.
Wispr Flow — Data Controls / help center · accessed Sep 2026 ↗
1Deletionstated
Account deletion is self-serve on desktop and iOS (Settings → Account → Delete account) and removes “dictionary, Flow Notes, preferences, transcripts, audio and usage history” locally and in the cloud, but Android has no in-app button and needs a support ticket. Some requests have a waiting period before final deletion and no purge timeline is stated; organization, privacy-request, audit and HIPAA records are kept, and local history on other devices is not removed.
Wispr Flow Help Center — Manage your Flow account · accessed Sep 2026 ↗
2Trains AIstated
Training is opt-in by the policy’s own framing — “if you opt to share your content with us for model training” via the “Improve the model for everyone” toggle — meeting audio is excluded “unless expressly disclosed and enabled”, and third-party LLM providers are “not used to train their models”, with data “generally deleted within 30 days” by providers. Does not train by default. One path sits outside the toggle: rating a dictation or sending feedback shares that session’s audio, transcripts and edits, which Wispr uses to “develop and train our AI models” — optional and per action, not a default.
Wispr Flow Privacy Policy — model training · accessed Sep 2026 ↗
Graded 2026-09-01 against the Wispr Flow Privacy Policy (last updated Aug 19, 2026), the Data Controls page and the Wispr Flow help center. Consumer surface = the Wispr Flow dictation and Notetaker apps (Mac, Windows, iOS, Android), operated by Wispr AI, Inc. The stated privacy design is genuinely differentiated: model training is opt-in via the “Improve the model for everyone” toggle (formerly Privacy Mode), cloud storage of transcripts is a user choice, meeting audio is excluded from training “unless expressly disclosed and enabled”, and third-party LLM providers are barred from training on the data with roughly 30-day provider deletion. The rights mechanics lag the design: email-only requests with no stated windows, no export tool, and a cloud purge that support completes by hand. A young, fast-moving product — the policy is days old and help articles update weekly, so watch coverage carries more weight than usual. All cells stated, not tested.
Events
Breaches, policy changes, and export-format changes we’ve recorded at Wispr Flow — other parties’ actions, alongside the enforcement record above. An event never changes a grade by itself. All events →
1 Oct 2026Wispr Flow now discloses that feedback can train its models. When you rate a dictation or send feedback, it collects data from that session — which can include audio recordings, transcripts and your edits — and uses it to "develop and train our AI models." Staff and service providers may review it. Sending feedback is optional. The page also says Enterprise accounts have data sharing turned off, with no option to turn it on.policy change
On the events page →
25 Sep 2026Wispr Flow rewrote its account-deletion instructions. Deletion removes your dictionary, notes, preferences, transcripts, audio and usage history, locally and in the cloud. Some requests now have a waiting period before final deletion, and no purge timeline is stated. Wispr keeps organization records, privacy-request and audit records, and HIPAA agreements. Android still has no in-app deletion button; you file a support request.policy change
On the events page →
Grade history
Every change to this company’s grade since it was first published, and why. All grade changes →
5 Oct 2026Trains AI on your datanote
Score unchanged (2). Wispr Flow’s Data Controls page now says that rating a dictation or sending feedback shares that session’s audio, transcripts and edits, which it uses to train its models. This is optional and per action, so training is still not on by default (R4). The rationale now names this path. OpenAI’s Trains 1 rests on default-on training, not on its feedback path, so the two scores stay consistent.
Followed a published event →
5 Oct 2026Deletionnote
Score unchanged (1). The quoted sentence ("queued for a follow-up cleanup that support completes") left the help article in September. The rewritten article describes cloud deletion of transcripts and audio, a waiting period on some requests, and the records Wispr keeps. It still states no purge timeline, so R2 keeps the score at 1. The rationale now quotes the current text.
Followed a published event →