Dictation App Ratings: Average Stars on the App Store

Most voice and dictation apps on the App Store are rated near the top of the scale, so a high average tells you little on its own. The scores that matter are the exceptions. Legacy professional dictation software and Apple's built-in assistant sit far below the field, while newer apps built on modern speech models cluster near the top. The dataset holds 326 data points ¹, each one an app's average in one country's store, read from Apple's lookup API on the day of collection.

Newer apps cluster near the top of the scale

On the US App Store, the leaders are close to the ceiling. Granola, a meeting notes app, averages 4.96 stars ². Plaud averages 4.88 stars ³. ChatGPT voice mode reaches 4.83 stars ⁴, and it does so across a very large number of ratings. Among dictation apps, Wispr Flow averages 4.84 stars ⁵.

The chart below shows how the apps compare.

Scores this compressed have a practical consequence. A gap of a tenth of a star between two top apps is often within what chance and rating prompts can produce. The gap between Granola and PowerMic Mobile (2.62 stars ⁶) is not.

Legacy professional dictation is rated far lower

The clearest weak spot is the older professional category. Dragon Anywhere averages 2.58 stars in the United States ⁷, 2.34 in the United Kingdom ⁸ and 2.15 in Germany ⁹. PowerMic Mobile, from the same seller, averages 2.62 stars in the United States ⁶. Microsoft's newer Dragon Copilot sits at 3.61 stars ¹⁰, better but still behind the field.

The newer clinical documentation apps score much higher. Heidi Health averages 4.74 stars in the United States ¹¹ and Accurx Scribe 4.83 ¹². Suki (4.12 ¹³) and DeepScribe (4.03 ¹⁴) sit in between. Here is the United States view of the whole segment:

Showing 8 of 12 rows. See every row in the dataset.

The pattern is consistent with a split between apps people use as a separate dictation product and apps that are part of a larger workflow. The data does not say why. Reviews could reflect accuracy, price, subscriptions or app stability, and a star average cannot separate them.

Siri is the lowest-rated mainstream voice tool

Apple's own assistant averages 3.40 stars in the United States ¹⁹ and 2.69 in the United Kingdom ²⁰. Compare that with Amazon Alexa at 4.54 in the United States ²¹ and Google Assistant at 4.73 ²².

Siri's score needs a careful reading. It comes from a small number of ratings compared with the big assistants, and most people never rate a system feature. The people who do may be the ones with a complaint. The dataset also has no row for the dictation feature built into iPhone keyboards, so "built-in dictation" is only represented through this assistant app.

The same app can score differently by country

Country matters more for some apps than others. Gboard averages 4.06 stars in the United States ²³ but 3.20 in Japan ²⁴, a large gap with a large number of ratings behind each figure. Microsoft SwiftKey shows a smaller version of the same split: 4.58 in the United States ²⁵ against 4.11 in Japan ²⁶.

Just Press Record is another example. It scores 4.51 in Germany ²⁷ and 3.52 in Japan ²⁸. Microsoft Copilot goes the other way, with 4.87 in Brazil ²⁹ and 4.44 in Japan ³⁰.

Language is a plausible cause for keyboards and dictation, since recognition quality differs by language. But storefronts also differ in how often an app asks for a rating and in how users score. The data shows the gap. It does not show the reason.

Transcription apps vary more than dictation apps

Transcription apps show the widest spread among the categories with many rows. Notta averages 4.06 stars in the United States ³¹ and 3.89 in Germany ³². Temi averages 3.91 ³³. Whisper Transcription is at 3.90 ³⁴. In the same segment Riverside reaches 4.74 ³⁵ and Transkriptor 4.71 ³⁶.

Dictation apps for the desktop are tighter. Superwhisper averages 4.38 stars in the United States ³⁷, and Wispr Flow is at 4.84 ⁵. These are companion apps to products that run mainly on a computer, so the iPhone rating covers only part of what people use.

Where the data disagrees with itself

Meeting notes apps are mostly rated very highly, yet one of them is a clear outlier. Limitless averages 3.30 stars in the United States ³⁸ and 1.94 in Japan ³⁹, while Granola, Plaud and Otter.ai (US 4.79 ⁴⁰) sit near the top. A single segment label therefore hides both the best and the worst scores.

Perfect scores need caution. Monologue shows 5.00 in Brazil ⁴¹, and Utter shows 5.00 in the United States ⁴². These come from a very small number of ratings, only just above the minimum for a row. The same is true of low outliers such as Trint at 2.80 ⁴³. The table lists the number of ratings beside every average, and you should read it before you trust a score.

What the data cannot tell you

• Only the iPhone and iPad store has rows. There are no Mac rows, so Mac-only dictation apps and the large Mac market are missing. Browser extensions, speech-to-text APIs and open models have no rows either, because they are not App Store apps.
• Ratings are not accuracy. They reflect price, subscriptions, ads, reliability and how often an app asks for a rating. Apple can also reset an app's ratings when a new version ships, and the lookup can lag the store page by a day.
• Only a few people rate. Ratings are a small fraction of downloads, and they lean toward people who are very pleased or very annoyed.
• Storefronts with a handful of ratings are excluded. A small app can therefore be absent from a country where its rating would not be meaningful.
• It is one day's reading. The dataset is tracked weekly, so averages will move, and an app that fixes a problem can recover slowly because old ratings stay in the average.

What this means if you are choosing how to write

Use ratings to rule out, not to rank. A score in the low range is a real warning: it appears for Siri, for Dragon Anywhere and for Limitless, and the pattern holds across countries. A score in the high range tells you the app is not broken, nothing more.

Then check three things the stars cannot show. First, whether the app covers your language, since Gboard's and SwiftKey's scores fall in Japan. Second, whether your use is dictation, transcription of recordings or meeting notes, since these are different products with different failure modes. Third, whether you need the app on a computer, which this dataset does not cover. Our guide to dictation software compares the options by task, and the market overview shows who builds what.