How to Compare a Writer Draft Against Your Style Guide

Short answer: Run the draft through Writer’s review check with the style guide attached, filter flagged terms by severity, then accept or reject each suggestion one at a time. Do not bulk-accept — roughly a third of terminology flags are false positives on product names and quoted customer language.

This is the workflow question teams ask after week one, when the glow wears off and the flagged suggestions start feeling like noise.

How does Writer’s review check work?

The check runs two things in parallel: a dictionary pass against your terminology lists, and a model pass on voice traits. The dictionary pass is deterministic. The model pass is probabilistic, so the same paragraph can score differently across runs.

Because of that split, treat the two outputs differently. Term flags are almost always worth acting on. Voice suggestions are opinions — sometimes good ones.

What is the step-by-step review process?

1. Open the draft in the Writer add-on inside Google Docs or Word.
2. Confirm the correct style guide is selected in the sidebar. A draft reviewed against the wrong website’s guide produces confident nonsense.
3. Click Review and wait for flags to populate. Longer documents take 10–30 seconds per section on the enterprise tier.
4. Work through Terminology flags first, top to bottom. Reject any flag on a proper noun, a quoted customer quote, or a code snippet.
5. Then read the Voice suggestions. Accept rewrites that shorten a sentence; skip ones that flatten your point into neutral business prose.
6. Re-run the review once to catch flags introduced by your own edits. A single pass misses cascading changes.

That last step is where most quality comes from. Two passes catch nearly everything; five passes make the text worse.

Which flags should you ignore?

A practical triage list, built from running this on about 40 articles:

Flag type	Act?	Reason
Banned term in body copy	yes	that is the point of the list
Term in a direct quote	no	editing quotes damages credibility
Passive voice in a step instruction	sometimes	fine in procedural text, bad in a claim
Sentence too long (>35 words)	usually	long claims bury the point
“Unclear pronoun” on technical nouns	often no	model misreads product names as pronouns

Keep a shared note of the false positives your team rejects repeatedly. It becomes the argument for tightening the rule, not the model.

How do you score a draft quickly?

For a batch of ten freelance submissions, section-by-section review is too slow. Use the API instead and check a numeric score plus the flag list, then only human-review documents below your threshold.

The API returns flagged terms with positions, which lets you script a check that fails a draft if a hard-banned term appears in the first 200 words. That single rule catches more real problems than a full voice review, in my experience — opening paragraphs are where bad habits cluster.

Does the review check work on content you did not write in Writer?

Yes. It is a checker, not just an assistant. Paste text generated in another tool, or route it through the API, and the terminology pass works identically. Voice scoring is calibrated against your guide, so it reflects your rules rather than the source tool’s defaults.

That makes Writer usable as a gate in front of publishing even if your writers use something else. It is a legitimate reason to buy it — arguably more defensible than buying it as a drafting tool.

What are the limits?

• False positives on proper nouns. Product names inside banned-word lists get flagged every time unless you add them as allowed terms.
• Long documents. Past about 800 words, voice adherence degrades and the second half needs a manual read.
• No version diffing. You cannot compare a draft’s score to a previous revision in the UI; the API is the only way to track it over time.
• Cost. Enterprise quotes, annual billing, and the checker consumes the same seat as the writer.

Who should run this process?

Content teams with a compliance review step, or agencies delivering to a brand with a published style guide. A solo writer with a personal voice guide will find the UI overhead worse than a plain prompt.

One honest recommendation: if you only need the checker, test whether a cheaper tool with a custom dictionary covers your needs before committing to an annual enterprise contract. The value is in the rule enforcement, and simpler tools do dictionary passes well.

Quick Summary

• Process: review terminology flags first, then voice, then re-run once.
• Act on: banned terms in body copy, overlong sentences, passive voice in claims.
• Ignore: flags inside quotes and on product proper nouns.
• Batch option: use the API score plus a hard-banned-term rule on the first 200 words.
• Main limits: false positives, drift past 800 words, no UI version diffing.

Last updated October 2026.