Why is Apple Dictation so bad?

Apple Dictation feels bad mostly when the conditions are poor. It struggles with a distant or noisy microphone, with a sentence that switches between languages, and with names and technical words that a general model has rarely seen. The one independent measurement we found put Apple's speech API at a 10.3 percent word error rate on a single recording ¹, measured through a developer interface rather than the Dictation key. That is useful evidence, but it is one test, not a rating of every session.

The sound that reaches the model is the first thing to fix

Speech recognition is most sensitive to its input. In the 2017 Ravanelli et al. figures, a single distant microphone gave a word error rate about 3.2 times the close-talking rate ². The authors state that distant speech recognition is dramatically reduced compared with the close-talking case ³, and that it still lacks robustness when non-stationary noise and reverberation are present ⁴. Those conditions are not identical to a Mac on a desk, so read the 3.2 ratio as a rough guide, not a fixed number.

Apple's own Dictation help points at the same causes. It advises you to position yourself as you normally use your Mac, speak clearly at a volume that is neither too quiet nor too loud, and avoid background noise ⁵. For a noisy office or a room with echo, Apple says a headset microphone may help ⁶.

Mixed languages and missing vocabulary cause visible errors

Apple describes switching between languages, not recognizing two languages inside one sentence ⁷. If you dictate a sentence that moves between English and Spanish, expect errors, because Apple's pages describe switching between languages, not mixing them. Check the language setting first, then dictate one language at a time.

Vocabulary is the second cause. The Dictation pages we read describe no custom word list, so names, product names and jargon are where a general model is most likely to fail. A 2022 study of UK Supreme Court hearings reports that adding an in-domain language model and common legal phrases to a generic speech model can improve its word error rate ⁸. That study used a different engine, but it shows why domain words help.

The engine's own share of the error is not published

Apple does not publish a word error rate for the Dictation key, so the part of the problem that comes from the engine cannot be measured from Apple's pages. The 2025 9to5Mac test gave three results on the same recording: 10.3 percent for Apple's speech API ¹, 1.5 percent for Whisper Large V3 Turbo ⁹, and 12.3 percent for NVIDIA's Parakeet v2 ¹⁰. The author reached Apple's API through an open-source command-line tool, so the test measures the developer interface, not the Dictation key ¹¹.

Speaker differences also show up in older tests. In a 2020 study of commercial speech recognition, Apple's system had a word error rate of 0.45 for Black speakers and 0.23 for white speakers ¹². The gap is a finding about the systems tested in that study. Because the study was published in 2020, it cannot show how the current engine performs for any speaker.

Is it getting worse? We found no measurement that shows it

Some people report that dictation got worse after a macOS upgrade. We found no independent measurement of a regression, and Apple has not said that accuracy fell. The vendor-written reviews we found in 2026 disagree with one another, and most come from companies that sell competing dictation apps, so we leave their figures out.

The newest Apple model is limited. Apple lists the "More advanced Dictation" model in macOS 27 for 9 English variants ¹³, and it needs a Mac with M3 or later and at least 12GB of unified memory ¹⁴. If your language or Mac is outside that list, that model does not change what you hear.

It depends: when Apple Dictation is good enough

• Good enough: short English messages, notes and first drafts, dictated in a quiet room with a close microphone.
• Poor fit: long technical, legal or medical text, code, dense names, noisy places, and sentences that mix languages.
• Not a fit for a fixed accuracy target: Apple publishes no rate, and the one test we found is a single recording.

What to do: fix the input first, then the vocabulary

1. Move the microphone close and quiet the room. If you use an external microphone, make sure its input volume is sufficient ¹⁵. The microphone guide covers the choice in more detail: microphones for Mac dictation.
2. Set one dictation language and dictate one language at a time. You can add languages in Dictation settings with the Edit button next to Languages ¹⁶.
3. Measure before you switch. Dictate a 200-word passage of your own writing, count the words that came out wrong, and divide by 200. Repeat with your usual microphone and the built-in one. The question on whether Apple Dictation is any good explains how to read the result.
4. Add names and jargon through Voice Control if you need them. Voice Control accepts up to 1,000 vocabulary terms ¹⁷. Turning it on replaces standard Dictation ¹⁸, so set it up for the work that needs it. The question on adding words has the steps.
5. If the test still fails, try a dedicated app. Yap, which we make, is an on-device dictation app for Mac and Windows. Test it with the same passage and microphone you used for Apple's Dictation, and compare the counts yourself.

The overview of how Apple Dictation works covers where the speech model runs, and the question on whether it learns your voice explains why it has no training step. For the settings that affect accuracy in more detail, read how to improve dictation accuracy, and for a step-by-step fix of a Mac that will not type at all, see why Mac dictation is not working. The word error rate entry explains how the percentages are calculated.