Editing for Voice vs. Editing for Accuracy: They Are Not the Same Pass

The instinct when editing an AI draft is to fix everything you notice in one pass. A factual claim you are not sure about, a sentence that sounds wrong, a hedging phrase that needs to go, a section that makes an argument you want to verify. You move through the document addressing each problem as it appears.

The problem is that accuracy checking and voice editing require different modes of attention. Combining them means you catch some of each and miss most of both.

What each pass is looking for

An accuracy pass is an analytical operation. You are asking: is this claim true? Is this source reliable? Is this statistic current? Does this characterization match reality? You are reading in a skeptical, questioning mode, holding the draft at arm's length and probing it.

A voice pass is a reading operation. You are asking: does this sound like me? Does this sentence have the right rhythm? Is the register pitched at the right level of specificity? You are reading as a reader would read, which requires being inside the piece, not skeptical of it.

These modes are not compatible. Being skeptical of the claims interrupts the experience of reading the prose as a reader would. Being inside the prose as a reader prevents the analytical distance required to question the facts. When you attempt both simultaneously, you are context-switching constantly, which reduces the effectiveness of both.

Running them separately

The accuracy pass comes first. Read the piece as a fact-checker would read it. Flag every factual claim, every statistic, every attribution. Do not edit the prose. Do not change words for clarity or flow. Your only job is to identify everything that could be wrong and verify it.

Once the accuracy pass is complete and the facts are right, run the voice pass. Now you read as a reader. You are looking for the AI tells, the flat passages, the hedging phrases, the structural symmetry that signals machine-generated text. You are also looking for places where the piece sounds generic rather than specific, where the argument loses specificity, where the register slips.

The sequence matters. Running voice first and then accuracy means you may edit a passage into better prose and then discover the underlying claim is wrong. You have now done the work twice: once to improve the voice, once to fix the fact and re-improve the voice around the correction.

Why this is easy to ignore

Running two passes takes longer than running one pass, or at least it feels that way. In practice, two focused passes are usually faster than one unfocused pass, because each pass is doing one thing with full attention.

The real resistance is that it requires discipline to resist fixing voice problems during the accuracy pass and vice versa. When you are reading an AI draft and you see a sentence that sounds clunky, the instinct is to fix it right now. The discipline required is to mark it and come back.

A simple method: use two different annotation colors or symbols in your editor. One for accuracy flags, one for voice flags. Read through once, flagging without editing. Then resolve each category of flag in a separate pass. The physical separation of the marking from the editing makes it easier to maintain the distinction.

What to do when the fact and the prose are tangled

Sometimes an accuracy fix requires rewriting the prose around it. The draft says something that is wrong in a way that cannot be corrected by substituting a different number or removing a phrase. The whole sentence needs to change because it was built around a false claim.

In that case, fix the fact first and write the replacement sentence as a placeholder without worrying about voice. Then, in the voice pass, revisit the placeholder and bring it up to the voice standard of the surrounding text. The sequence still holds: accuracy before voice, even when accuracy creates a prose problem.