Part 3: Define Your Brand Voice System
The Problem
AI content sounds like AI. Every post reads like it was written by the same enthusiastic intern who just discovered a thesaurus.
The solution isn't better prompts. It's a systematic voice ruleset that your agent enforces automatically — a validation layer that catches and rejects bad patterns before anything gets published.
Why 90% of AI Content Sounds the Same
Every large language model defaults to the same voice: enthusiastic, generic, packed with corporate jargon.
The Default AI Vocabulary
Ask any LLM to write a LinkedIn post and you'll get these words. Every time. The output is technically correct but completely forgettable.
Asking “write naturally” does not work
The AI interprets “natural” as “slightly less corporate.” You still get the same vocabulary with a few contractions thrown in.
Building a RULESET works
A validation layer that literally rejects posts containing forbidden words before they ever get published. Rules are enforceable. Vibes are not.
Jenny's Approach
Every post Jenny generates goes through automated validation. If the text contains a forbidden word, it gets sent back to the LLM with explicit instructions to fix the violation. No human intervention required.
Jenny's Brand Voice Rules (Steal This)
These are the actual rules that govern every piece of content Jenny produces. Adapt them for your brand.
Universal Rules — Apply to Every Platform
ZERO emojis
Absolute zero tolerance. Regex catches every Unicode emoji range. No exceptions.
ZERO hashtags
Exception: Instagram allows max 3 and must include the brand hashtag.
ZERO third-party links
Only your own properties. Every outbound link must point to a domain you control.
ZERO corporate jargon
Numbers always as numerals — 30%, not thirty percent
No rhetorical questions — state facts, don't ask vague questions to “engage”
1 sentence = 1 paragraph — short, scannable, punchy
Hook-driven opening — the first line decides if anyone reads the rest
Write as you speak — simple, plain, inclusive
Confidence over hedging — state facts, don't waffle
Platform-Specific Voice Tuning
The same article becomes 6 different posts. Each platform gets an “edginess scale” from 1 to 10 that controls how provocative the AI gets.
X (Twitter)
8.5/10Sharp, punchy, opinionated
Max 280 chars, shortenings OK
Professional but human
800-1200 chars, subtle sarcasm only
Visual-first, caption supports image
Max 3 hashtags, "link in bio" CTA
Threads
5.5/10Conversational, provocative
Keep SHORT: 400-600 chars
Maximum clickbait hook (without being misleading)
600-900 chars, pattern interrupts
Bluesky
7.0/10Clever, not corporate
300 chars total, internet-native
The Edginess Scale Concept
Each platform gets a number from 1 to 10 that controls how provocative the AI gets. This single number changes the entire character of the output.
LinkedIn at 3.0
Measured and professional. Data-driven observations. No hot takes.
X at 8.5
Bold claims and hot takes. Opinionated. Punchy. Designed to stop the scroll.
The Validation Layer — Your Quality Gate
Without automated enforcement, rules are just suggestions. The validation layer is what makes voice consistency possible at scale.
The Validation Loop
LLM generates post
Content is created based on source material and platform rules
Automated validation runs
Every check fires: emojis, hashtags, forbidden words, character limits, link policy
Pass
Post gets published
All checks passed. Content goes live.
Fail
Post goes back to LLM
Specific error instructions included in the retry prompt.
Loop continues until pass — or gets flagged for human review
Most posts pass on the 1st or 2nd attempt. Edge cases get escalated.
What the Validation Layer Checks
Emoji regex
Catches ALL Unicode emoji ranges — not just the common ones
Hashtag detection
Blocked on all platforms except Instagram (where max 3 are allowed)
Forbidden word scanning
Case-insensitive check against the full banned word list
Character limit enforcement
Includes URL shortening math for X — t.co makes every URL 23 chars regardless of original length
Link policy check
Only your own domain allowed. Third-party links get rejected automatically.
UTM parameter verification
Every link must have proper tracking parameters attached
Building Your Voice Document
Use this template to define your own brand voice system. Fill in each field and hand it to your agent as configuration.
Voice Document Template
Brand personality in 3 words
Forbidden words (list at least 10)
Emoji policy
Hashtag policy
Link policy
Default tone (1-10 scale)
Platform overrides
X → ___
LinkedIn → ___
Instagram → ___
Threads → ___
Facebook → ___
Bluesky → ___
What You Just Learned
- AI content sounds generic because there are no rules, not because the AI is bad
- Build a ruleset: forbidden words, tone scale, platform-specific adjustments
- Add a validation layer that automatically rejects bad output
- The validation loop (generate → validate → fix → re-validate) is what makes quality consistent
Next: Your voice system is defined. Now let's build the content pipeline that turns source material into platform-ready posts.