Antigravity Brand Voice Training Guide: Build AI Content That Sounds Like Your Team Wrote It

Why Brand Voice Is the Difference Between AI Slop and AI Content

Every company that tries AI content generation hits the same wall: the output is grammatically correct, factually reasonable, and completely generic. It reads like it was written by a competent stranger who has never met your customers, used your product, or attended your team meetings. This is the brand voice problem.

Antigravity solves this by allowing you to train its AI on your specific brand voice — not through fine-tuning a model, but through a combination of reference content analysis, explicit voice parameters, and terminology control. When configured properly, Antigravity produces content that your own team members cannot distinguish from human-written drafts.

This guide covers the systematic process for training Antigravity on your brand voice and maintaining consistency as your content operation scales.

Step 1: Audit Your Existing Brand Voice

Collect Your Best Content

Before configuring Antigravity, identify 15-20 pieces of content that best represent your brand voice. These should be:

• Published and approved: content that went through your review process and was deemed on-brand
• Diverse in format: blog posts, email newsletters, product descriptions, social media posts, help articles
• Recent: written within the last 12 months (voices evolve)
• High-performing: content that resonated with your audience (high engagement, shares, conversions)

Analyze Voice Characteristics

Read through your collected samples and document:

Tone spectrum:

Formal ←——————————→ Casual
  1    2    3    4    5
Our brand: 3.5 (professional but approachable)

Personality traits:

- Confident but not arrogant
- Helpful but not patronizing
- Technical but not jargon-heavy
- Opinionated but backed by data
- Warm but not overly familiar

Sentence patterns:

- Average sentence length: 15-20 words
- Mix of short declarative and longer explanatory sentences
- Occasional questions to engage the reader
- Active voice by default, passive only for emphasis
- Contractions used (we're, you'll, it's) — not overly formal

Vocabulary preferences:

We say:          We never say:
"customers"      "users" or "consumers"
"team"           "workforce" or "human resources"
"straightforward" "simple" (implies customer's problem is trivial)
"challenge"      "problem" (more positive framing)
"guide"          "tutorial" (too technical-sounding)

Step 2: Configure Brand Voice in Antigravity

Upload Reference Content

Upload your 15-20 reference pieces to Antigravity’s Brand Voice section:

1. Navigate to Settings > Brand Voice
2. Click “Upload Reference Content”
3. Upload files in order of importance (most representative first)
4. Add a brief description for each: “Blog post — technical topic, conversational tone” or “Product page — persuasive, benefit-focused”

Format recommendations:

• Blog posts: upload as clean text or markdown (remove navigation, footers, sidebars)
• Emails: include subject line and preview text
• Social posts: batch multiple posts into one document, grouped by platform
• Product pages: include headlines, subheadlines, and body copy — not just descriptions

Define Voice Parameters

Antigravity’s voice configuration panel lets you set explicit parameters:

Brand Voice Configuration:

Company Name: TechForward
Industry: B2B SaaS (project management)
Target Audience: Engineering managers and product leads at
  mid-market companies (100-1,000 employees)

Tone: Professional but warm. We sound like a smart colleague
  who genuinely wants to help, not a corporate manual.

Perspective: First person plural ("we") for company content.
  Second person ("you") for educational content.
  Never third person ("TechForward believes...").

Energy Level: Medium-high. We are enthusiastic about our product
  and our customers' success, but we do not use exclamation marks
  excessively or hype language.

Humor: Occasional dry wit. Never jokes at the customer's expense.
  Self-deprecating humor is acceptable in blog posts.
  No humor in help documentation or error messages.

Technical Level: Our readers are technical but busy. Explain
  concepts without being condescending. Use industry terms
  (sprint, CI/CD, deployment) without defining them. Define
  niche terms on first use.

Formatting Preferences:
  - Short paragraphs (2-4 sentences max)
  - Bullet points for lists of 3+ items
  - Bold for key terms on first mention
  - Headers every 200-300 words
  - No walls of text

Create the Terminology Glossary

Upload a glossary that controls how Antigravity uses your specific terms:

Terminology Glossary:

Product Names:
  - "TechForward" (always capitalized, one word)
  - "TechForward Teams" (our team plan — always with capital T)
  - "TechForward Enterprise" (enterprise plan)
  - Never: "Tech Forward," "Techforward," "TF"

Feature Names:
  - "Sprint Board" (capitalized, our Kanban feature)
  - "Flow Mode" (our focus time feature)
  - "Team Pulse" (our team health check feature)
  - Never abbreviate feature names

Competitor References:
  - Mention competitors by name only in comparison content
  - Never disparage — use factual comparisons only
  - Approved competitors for comparison: Asana, Monday.com, Linear
  - Do not compare to: Jira (different category)

Industry Terms:
  - "agile" (lowercase unless starting a sentence)
  - "DevOps" (capital D, capital O)
  - "CI/CD" (always with slash, always capitalized)
  - "standup" (one word, not "stand-up")

Banned Phrases:
  - "leverage" (use "use" instead)
  - "synergy" (use "collaboration" or "integration")
  - "cutting-edge" (use "modern" or "advanced")
  - "seamless" (overused — describe the actual experience)
  - "best-in-class" (prove it with data instead)
  - "in today's fast-paced world" (delete entirely)

Step 3: Test and Calibrate

The Blind Test Method

After configuration, generate 5 test pieces across different content types:

1. A blog post introduction (200 words)
2. A product feature description (100 words)
3. A customer email response (150 words)
4. A social media post (50 words)
5. A help article section (200 words)

Mix these with 5 human-written pieces of the same types. Ask 3 team members to identify which are AI-generated. If they consistently identify the AI content, the voice needs more tuning.

Common Calibration Issues

Issue: Too formal The AI defaults to corporate language. Fix: add more casual reference content. Explicitly note: “Use contractions. Write like you’re explaining to a colleague, not writing a press release.”

Issue: Too generic The content is correct but lacks personality. Fix: add more opinionated reference content. Include examples of your brand taking a stance: “We believe sprints should be two weeks, not one.”

Issue: Wrong vocabulary The AI uses competitor terminology or industry jargon you avoid. Fix: expand the terminology glossary with more “we say / we never say” pairs.

Issue: Inconsistent across formats Blog posts sound different from emails. Fix: create format-specific voice notes: “Blog posts are more conversational. Help docs are more direct. Social posts are punchier.”

Step 4: Maintain Voice Consistency at Scale

Content Type Templates

Create templates for each content type with voice-specific instructions:

Template: Blog Post
Voice notes: Conversational, opinionated, educational.
  Start with a hook that relates to the reader's daily work.
  Include at least one concrete example or data point per section.
  End with a practical takeaway, not a sales pitch.
Structure: Hook → Problem → Solution → Example → Takeaway
Word count: 1,200-1,800 words
Reading level: Grade 9-10

Template: Product Update Email
Voice notes: Enthusiastic but not hype. Focus on what the
  customer can now do, not what we built.
  Lead with the benefit, not the feature name.
  Include one screenshot or GIF.
Structure: What's new → Why it matters → How to use it → CTA
Word count: 200-400 words
Tone: Slightly more excited than blog posts

Quarterly Voice Review

Every quarter, review the brand voice configuration:

1. Collect feedback: ask your team which AI-generated content felt off-brand
2. Audit recent output: read 10 randomly selected AI-generated pieces
3. Update reference content: replace older references with recent high-performing content
4. Refresh terminology: add new terms, remove outdated ones
5. Re-run the blind test: verify calibration is still accurate

Team Onboarding

When new team members join the content team:

1. Share the brand voice configuration document
2. Have them read the 15-20 reference content pieces
3. Ask them to write a short piece and compare against AI output
4. Use discrepancies as a teaching tool for brand voice nuances

Measuring Brand Voice Consistency

Metrics to Track

Frequently Asked Questions

How many reference pieces do I need?

Minimum 10, recommended 15-20. More is better up to about 30 — beyond that, diminishing returns. Focus on quality and diversity of formats over quantity.

Can I have multiple brand voices?

Yes. Create separate voice profiles for different brands, products, or audience segments. A B2B product and a consumer product from the same company should have different voice configurations.

How often should I update the voice configuration?

Quarterly for minor updates (new terms, fresh reference content). Annually for major reviews (tone adjustments, audience changes). Immediately when you rebrand or significantly change positioning.

Does the brand voice affect all content types equally?

The voice parameters apply globally, but templates add format-specific adjustments. A help article and a blog post share the same vocabulary and tone spectrum, but the template controls structure and energy level.

Can Antigravity match a specific person’s writing style?

Yes, if you upload enough reference content from that person (10+ pieces). This is useful for ghostwriting or maintaining a founder’s voice across content written by the team.

What happens if my brand voice changes?

Update the configuration and reference content. Antigravity adapts immediately — there is no retraining delay. Old content is not affected; only new generations use the updated voice.