Build a brand voice QA system for AI-generated content

Gather 8-12 examples of content that sounds right: website copy, founder posts, product launch emails, customer stories, sales decks, and high-performing LinkedIn posts. Put them in Google Docs with notes on why each example works. Also collect 5 examples that do not sound right so the QA system learns what to reject, not just what to imitate.

Use Claude to analyze the examples and produce practical voice rules: sentence length, tone, vocabulary, structure, level of directness, proof requirements, and banned patterns. Convert the output into a simple one-page reviewer guide.

Analyze these brand voice examples and create a practical QA guide for AI-generated GTM content.

Approved examples:
{{approved_examples}}

Rejected examples:
{{rejected_examples}}

Brand context:
{{brand_context}}

Output:
1. Voice summary
2. What the brand should sound like
3. What the brand should never sound like
4. Sentence and structure rules
5. Words and phrases to prefer
6. Banned phrases and patterns
7. Claim/proof rules
8. 10-point QA checklist

Make this specific enough that a junior marketer can use it to review AI drafts.

Build a list of phrases, structures, and claims you want to catch automatically. Include generic AI phrases, overused openings, unsupported superlatives, filler transitions, vague transformation claims, and words your brand would never use. Store the list in Notion or Google Docs so it can be updated every month.

Use Claude to turn the voice guide and banned pattern list into a reusable QA prompt. The prompt should score drafts, flag claim risks, rewrite only the weak sections, and explain why each change is needed. Keep the prompt modular so reviewers can paste in blog posts, emails, ads, landing pages, or social posts.

Review this AI-generated GTM content against our brand voice QA guide.

Content type:
{{content_type}}

Target audience:
{{target_audience}}

Draft content:
{{draft_content}}

Brand voice guide:
{{brand_voice_guide}}

Banned phrases and patterns:
{{banned_phrase_library}}

Output:
1. Overall score from 1-10
2. Voice match score
3. Clarity score
4. Proof/claim risk score
5. Generic AI writing flags
6. Banned phrase flags
7. Specific sections to fix
8. Revised version with minimal edits
9. Final approval recommendation

Do not rewrite strong sections unnecessarily.

Before human review, run the draft through Writer or Grammarly to catch grammar, style consistency, readability, and terminology issues. Then compare those results with the Claude QA output. Use the tools as filters, not final authority, because brand voice is more nuanced than grammar quality.

Create a simple risk tier: low-risk social posts, medium-risk campaign copy, high-risk claims, legal/compliance content, and executive comms. Define who approves each type and where feedback lives. Add the checklist to Notion so the same review path is used every time.

Test the workflow on 5-10 recent AI-generated drafts. Track recurring issues, false positives, and reviewer disagreements. Update the banned phrase library and QA prompt based on what actually shows up in drafts.