Should you edit while you dictate or afterward?

Edit afterward, in a separate pass, rather than correcting as you speak. Correction takes most of the task time in dictation, and a correction spoken aloud is often misrecognized as well. Keep speaking, let the draft run to the end of the section, and then read it once and fix it. The exception is an error that would throw off every sentence after it, such as a wrong name that recurs or a missing negative that reverses the meaning. Fix those at once, with the simplest command your tool offers.

This is the same advice the guide on how to dictate gives in its section on two separate passes. The reasons are the research on how people split their time between dictating and correcting, and the research on what happens when they try to fix errors by voice.

Correction takes most of the time in dictation

When people dictate with a speech recognizer, most of the time goes to fixing errors, not to speaking. In a large-vocabulary study, correcting errors "dominates task time" ¹. A preliminary 2001 study found that users spent over 65 percent of their time correcting and less than 35 percent dictating ². Those studies are old, so the exact share will differ today. The direction is consistent, though: correction is the expensive part.

Stopping to correct each error as it appears spends that expensive time at the worst moment. You break the thought you were speaking, and you lose the sentence you were planning. A draft written in one pass has fewer breaks, and the correction happens once, when you can see the whole paragraph.

Spoken corrections often fail too

Many people try to fix a misheard word by saying the word again. This often fails. Research on dictation corrections found that the spoken correction is itself frequently misrecognized, which costs time and tests the user's patience ³. That was a 2006 study, so current tools may do better, but the lesson holds for re-speaking a phrase.

Novice dictators are the most likely to keep fixing errors by voice. Kraal and colleagues say that persisting with spoken error correction shows an incorrect mental model of dictation ⁴. Their observation is an argument from experience, not a controlled test, but it matches the time costs above. Users who keep dictating for about 20 hours start to mix voice correction with keyboard and mouse correction ⁵.

In a 1999 study, correction that combined speech with typing was faster than speech-only correction, but for users with good typing skills, correction by typing alone was faster still ⁶. That study had a small sample and an old recognizer, so read it as a pointer, not a rule.

Read the draft, then edit by keyboard

The usual order is three steps:

1. Dictate the section. Keep the rhythm steady, and do not stop at each error. Mark anything you are unsure of by saying it clearly and moving on.
2. Read the whole section once. Look for names, numbers, negatives and words that sound alike and are spelled differently.
3. Edit with the keyboard or with a single command. Use the keyboard for anything longer than a word, and a command for simple removals.

Many dictation tools have commands for these edits. In Dragon 15, saying "Correct That" opens the Correction menu for the last text you dictated ⁷. In Windows voice typing, commands such as "delete that" and "select that" edit text ⁸. Use these for quick fixes right after a sentence. For a whole paragraph, the keyboard is usually faster.

When to fix immediately

There are cases where waiting costs more than stopping:

• A wrong name or term that recurs. If the software writes the wrong name for a person, every later sentence will repeat it. Correct it once with a command, or add the word to your tool's vocabulary.
• A missing "not" or "no". A reversed meaning in an early sentence can lead you to dictate the rest on a false assumption. Fix it before you continue.
• A structural mistake. If you spoke a whole paragraph in the wrong place, moving it is easier now than later.

Everything else can wait. Small spelling and punctuation errors are easier to see on a full read, and a fix made in a read-through is usually cheaper than a fix made mid-sentence.

Automatic cleanup changes the answer

Some tools now correct the draft as you speak. Microsoft describes Fluid dictation, a Voice Access feature on Copilot+ PCs, as correcting "grammar, punctuation, and filler words as you speak" ⁹. With a feature like that, fewer errors reach the page, so the read-through is shorter. It does not remove the need for a read, because the tool cannot know what you meant.

Experiments with spoken editing show that the way you phrase a fix matters. In a 2021 study of a voice assistant used without a screen, the authors' editing techniques lowered the average failure rate by 54.1 percent compared with descriptive commands or re-speaking ¹⁰. That study used an eyes-free phone assistant, not desktop dictation, but it supports a simple habit: say the word you want to change, not a description of where it is.

Learning the balance takes practice

Early on it is tempting to edit too much during dictation, because the errors are visible and the urge to fix them is strong. Over time the balance shifts. Feng and colleagues found that users became more efficient at dictation, navigation and error correction as they gained experience ¹¹, and that navigation and correction take a large share of the work in the early stages ¹². The skill to build is restraint: keep going, and save the fixes for the read-through.

For the details of voice editing commands, see dictation commands, and for the errors themselves, see dictation mistakes. If you want to know how to check a draft, the question how to proofread dictated text covers the read-through in more detail. For the same method applied to letters and notes, read how to dictate a letter and how to dictate notes quickly. The term voice typing is defined in the glossary.