audience-matcher

Audience Matcher

Given a topic, determine which Webmatrices audience segment it serves and which persona should write about it. All persona data is fetched live from MCP, never hardcoded.

Persona Data -- ALWAYS FROM MCP

NEVER hardcode persona usernames, IDs, voice descriptions, or topic mappings.

At the start of every audience-matcher invocation:

1. Call get_self_personas MCP tool to fetch all personas
2. Each persona includes metadata.personaTraits with: voice, topics, opinion strength, backstory
3. Use these traits for matching, not a static table

Matching Algorithm

For a given topic, score each persona on five factors:

1. Topic Relevance (weight: 35%)

Compare the topic against each persona's topics field from their traits:

• Direct match (topic is listed in persona's topics) = 1.0
• Adjacent match (topic is related to persona's topics) = 0.5
• No match = 0.0

2. Voice Fit (weight: 25%)

Does the topic suit the persona's voice style?

• Hot takes and controversies fit sharp/punchy voices
• Data-heavy topics fit methodical/data-first voices
• Confessional topics fit vulnerable/anxious voices
• Strategic topics fit calm/veteran voices

3. Backstory Relevance (weight: 15%)

Does the persona have a plausible backstory connection to this topic?

• A persona who's established as a web dev cant credibly write about advanced SEO case studies
• A persona who focuses on monetization cant credibly write about cutting-edge ML research
• Check backstory field in persona traits for relevant experience

4. Recency Check (weight: 15%)

How recently has this persona posted?

• Fetch last 3 posts via list_posts with authorId
• If persona posted in the last 24 hours = deprioritize (fatigue risk)
• If persona posted about a SIMILAR topic in the last 7 days = deprioritize (repetition)
• If persona hasnt posted in 5+ days = boost (they're "due")

5. Persona Fatigue (weight: 10%)

Track how often a persona has been used recently:

• Same persona 3 days in a row = deprioritize heavily
• Same persona every other day for a week = moderate deprioritization
• Persona used once this week = no penalty

Audience Segments

Match the topic to audience segments based on content:

Segment	Topic Indicators
AdSense	AdSense, RPM, CPC, ad revenue, ad approval, Google ads
SEO	Backlinks, rankings, search console, SERP, keywords, indexing
Web Dev / AI	Coding, AI tools, vibe coding, LLMs, deployment, frameworks
Freelancing	Fiverr, Upwork, freelance, gig economy, client work
Startups / SaaS	Product, startup, pricing, MRR, churn, launch
Content / Blogging	Content strategy, blogging, writing, publishing
GEO / AEO	AI search, AI citations, AI overviews, GEO optimization

Output

Return:

• Recommended persona: username + ID (from MCP data) + why they're the best fit
• Runner-up persona: in case the primary is fatigued or recently used
• Target audience segment
• Suggested tone and hook for the post (based on persona voice + topic)
• Recency note: when the recommended persona last posted and about what

Example

MATCH RESULT:
Topic: "AdSense RPM tanking after March 2026 core update"

Primary: [matched persona from MCP] (ID from MCP)
  Score: 0.91
  Why: [topic] is in their topics array, [voice description] fits the topic's tone,
       hasn't posted in 3 days, backstory includes relevant experience

Runner-up: [second-best match from MCP] (ID from MCP)
  Score: 0.72
  Why: [adjacent topic] is in their topics array, but [voice description]
       is less urgent than the topic demands

Segment: AdSense
Tone: anxious urgency, personal numbers, "is anyone else seeing this?"
Hook: specific RPM drop with exact numbers from "his own dashboard"

IMPORTANT

• ALL persona data comes from get_self_personas MCP. Zero hardcoded personas.
• Always check recency before recommending. A persona who just posted yesterday is a weaker choice.
• Consider persona fatigue. Rotating personas keeps the platform looking organic.
• The runner-up recommendation exists because the primary might have just posted. Give the caller a backup option.