Back to the blog

APP STORE REVIEW STUDY

We read 1,281 App Store reviews of AI note takers for privacy and bot complaints

Hand-coded study of recent US App Store reviews for Otter, Granola, Fireflies, Plaud and Fathom: privacy, bot and consent complaints, with method and limits. Aurora is not in the sample.

Published · Updated

What did we find in recent App Store reviews?

On 6 November 2026 (Europe/Warsaw), we sampled 1,281 US App Store reviews. Hand-coded complaints appeared in 1.4% of Otter's sample, 3.4% of Fireflies' and 1.1% of Granola's (account/identity, not bots). Plaud and Fathom had none coded. Aurora was not in the sample. These are sample shares, not rates of real-world harm.

How was the review sample collected?

The source was Apple's public iTunes Customer Reviews RSS, using the US storefront, sortby=mostrecent, and pages 1 to 10. App track IDs were resolved with the iTunes Search API on 6 November 2026. The collection retrieved 500 Otter reviews, 187 Granola reviews, 88 Fireflies reviews, 500 Plaud reviews and 6 Fathom reviews.

For each app ID, the collection requested the feed pattern https://itunes.apple.com/us/rss/customerreviews/page={n}/id={id}/sortby=mostrecent/json. The table links each app's first feed page. A current feed can change and need not reproduce the historical sample.

US storefront collection, 6 November 2026
App and RSS sourceTrack IDReviews obtainedCollection limit
Otter Transcribe Voice Notes1276437113500RSS cap reached
Granola: AI Meeting Notes6739429409187RSS ran out
Fireflies: AI notetaker646316420388RSS ran out
Plaud: AI Work Companion6450364080500RSS cap reached
Fathom AI Notetaker67516134026All RSS returned

Which reviews counted as complaints?

The first pass matched title plus body against case-insensitive patterns with word boundaries. The filter covered these terms and phrases; encrypt* and upload* include word variants:

  • privacy, private, confidential, confidentiality, hipaa, gdpr, secure, security, encrypt*
  • bot, bots, notetaker joined, note-taker joined, uninvited, stranger joined
  • consent, without permission, didn't ask, recorded without, without telling, secretly record
  • train on, training data, my data, sell data
  • upload*, in the cloud, cloud server, data leak, leaked, spy, spying, snoop

Every flagged review was then read. A hand-coded complaint required an objection to privacy, consent, secret or unwanted recording, a meeting bot joining, unwanted sharing, or a security incident. A keyword hit alone did not qualify. These are the collector's recorded hand-coded totals, not totals produced by rerunning a keyword search.

False positives included feature requests about file uploads, praise using the words secure or bot, billing complaints about uploading a payment card, and unrelated animal or food language. For example, a Plaud reviewer praising support as not being a bot was not complaining about meeting bots.

The distinction matters: Otter had 18 keyword hits but 7 hand-coded complaints; Plaud had 7 keyword hits but no hand-coded complaints. Publishing keyword hits alone would overstate the result.

What did the hand-coding count?

The table reports the historical sample, not every review of these apps. Percentages are rounded sample shares. Granola's two account/identity complaints are not meeting-bot complaints. Fathom's six reviews are too few for a meaningful comparison.

Hand-coded complaints in the 6 November 2026 US RSS sample
AppSampleHand-coded privacy / bot / consent / security complaintsOf those, 1 or 2 stars
Otter5007 (1.4%)6
Granola1872 (1.1%, account/identity)2
Fireflies883 (3.4%)2
Plaud50000
Fathom600

What were reviewers objecting to?

  • Otter: four coded complaints concerned a bot or recording without knowledge or consent; three concerned privacy, security or sharing incidents.
  • Granola: both complaints concerned account/identity providers, including a required Google account and no Sign in with Apple. None concerned a meeting bot. Two keyword hits praised Granola for not joining as a bot.
  • Fireflies: coded complaints described an always-on microphone, required calendar access, and meetings shared with a company or clients who attended only part of a call. A separate mixed four-star review praised privacy but described difficulty stopping meeting joins; it is not an additional complaint in the headline total.
  • Plaud: every keyword hit was a false positive under the coding rule, including support described as not being a bot and uploads discussed as a feature.
  • Fathom: there were no coded complaints in the small sample. That does not establish an absence of privacy problems.

These descriptions report what reviewers alleged. We did not independently establish that the reported behavior occurred or that it describes the products' current settings.

What did the public reviews actually say?

These short excerpts are public App Store text preserved in the collected matching reviews. Dates and stars belong to those reviews, not to a new test of the apps. Each quote is evidence of a complaint, not proof of current product behavior.

The bot automatically joining my meetings, even after deleting the app was pretty creepy.

Otter, Creepy Bot, 1 star, 14 April 2025 (Otter App Store reviews).

It can appear that you are sending an AI agent to spy on meetings.

Otter, For experts only, 1 star, 12 September 2025 (Otter App Store reviews).

When using fireflies talk, even when I am not dictating, the mic is always on. Privacy concerns and annoying

Fireflies, Mic always on, 3 stars, 23 October 2026 (Fireflies App Store reviews).

The current App Store links may reorder or stop showing older reviews. The collection date remains 6 November 2026; publishing this article does not turn that sample into a fresh collection.

What can this study not tell us?

  • It is not a census. The sample is recent US iOS reviews returned by RSS, with caps for Otter and Plaud. The reviews themselves span different dates and app versions.
  • It is not the whole privacy audience. People who object most strongly to this kind of processing may never install an app or leave an App Store review.
  • It is not a rate of real-world harm. Reviewers self-select; the denominator is reviews retrieved, not users, meetings or recordings. The percentages do not rank product safety.
  • It is not an audit of current defaults. A public complaint does not establish the event, its cause, or what a current build does.
  • Fathom is not comparable here. Six reviews are too few to interpret. Zero coded complaints does not mean zero risk.
  • Keyword hits are not complaints. The filter can flag praise or feature requests, and this sample does not establish how many relevant complaints the filter missed.

Why is Aurora publishing this?

Aurora is being built around a different speech-processing boundary: it does not join calls as a meeting bot, and capture, transcription, speaker separation and summaries run on the iPhone. After the first-run model download, that speech workflow runs on the phone without a connection. This is not a claim that the app is network-free; the Privacy Notice names its other network flows.

Aurora had no App Store listing at the time of collection and was not included in the sample. This study measures neither Aurora's users nor its accuracy or safety. No bot also means no automatic announcement to the room: asking people before recording remains your responsibility.

For the practical questions, read Private AI note taker: 8 questions to ask before you record and why Aurora records from the room. For product differences, see the Otter AI alternative comparison and Fireflies alternative comparison.

AURORAPRIVATE AI FOR CONVERSATIONS
TRANSCRIPTS MADE ON DEVICE

AI notes,
on device.

Aurora is an app for meetings, interviews and lectures on iPhone (including iPhone Duo), iPad and Mac. It transcribes the conversation live, tells the speakers apart and writes the summary and action items on the device. No meeting bot joins, and no recording is uploaded to transcribe it.

Private beta, by invitation