pf-review-responses

Turn a 2-hour weekly review-response chore into a 15-minute workflow — on-brand, personalized drafts for every review sentiment, with your data staying local.

Installation

1. Download the pf-review-responses.plugin file
2. Open Claude Desktop and navigate to Settings > Plugins
3. Click Install Plugin and select the downloaded .plugin file
4. The plugin will be installed and available immediately

Note: All data stays local on your machine. No external API calls or cloud storage required.

Why This Exists

Small ecommerce businesses and local shops receive customer reviews on Google, Amazon, Yelp, Etsy, and other platforms — but writing thoughtful, brand-consistent replies takes significant time. Existing tools like Podium ($4,788-7,188/yr) and Birdeye ($3,588+/yr) bundle expensive enterprise features most SMBs never use. This plugin handles the part that matters most: drafting responses using professional reputation management frameworks.

Quick Start

1. Run /review:review-setup — enter your business name, industry, and preferred tone (takes 2 minutes)
2. Run /review:review-load — paste reviews or upload a CSV export from any platform
3. Run /review:review-respond — get a DOCX with personalized draft responses using SERVICE/LAST frameworks
4. Copy responses into each platform — review and personalize as needed
5. Run /review:review-report at month-end for a benchmark-compared trend analysis
6. Run /review:review-analyze anytime for a competitive health check with NPS

Commands

Command	Description
/review:review-setup	One-time setup: brand profile (Keller's Prism), industry benchmarks, workspace
/review:review-load	Load reviews with ABSA sentiment, Plutchik emotion, HEARD triage
/review:review-respond	Generate responses using SERVICE/LAST models with FTC compliance
/review:review-report	Monthly report with NPS, benchmarks, weighted sentiment, risk alerts
/review:review-analyze	Competitive benchmark analysis with NPS and revenue impact estimates
/review:review-build	Full pipeline: load + respond in one command
/review:review-status	Show workspace status and output files

How It Works

1. Setup — Captures brand identity using Keller's Brand Identity Prism; loads industry-specific benchmarks from BrightLocal 2024 and ReviewTrackers 2024
2. Load — Reviews analyzed with Aspect-Based Sentiment Analysis (per-topic sentiment), Plutchik emotion classification (8 emotions × 3 intensities), and HEARD escalation triage (4 levels)
3. Respond — Claude drafts responses using SERVICE recovery model (negatives), LAST technique (moderate complaints), and Plutchik emotion-matched tone — with platform-specific formatting and FTC 16 CFR Part 255 compliance checks
4. Report — Aggregates monthly data with NPS estimation (Reichheld methodology), ABSA weighted sentiment scoring, industry benchmark comparison, and revenue impact estimates
5. Analyze — Competitive positioning against industry norms with SWOT analysis and prioritized recommendations

AI-Powered Features

This plugin leverages Claude's AI capabilities with professional domain frameworks:

• Aspect-Based Sentiment Analysis (ABSA) — Analyzes sentiment per topic (product quality positive, shipping negative) with 10-category taxonomy and weighted scoring
• Plutchik Emotion Classification — 8 primary emotions × 3 intensity levels drive response tone matching (anger→de-escalate, sadness→empathize)
• SERVICE Recovery Model — Structured Sorry→Expedite→Recover→Value→Inform→Compensate→Evaluate framework for all negative reviews
• LAST Technique — Listen→Apologize→Solve→Thank for moderate complaints
• HEARD Escalation Triage — 4-level protocol (Standard→Priority→Escalation→Crisis) from Disney Institute service recovery model
• NPS Estimation — Net Promoter Score from star ratings using Reichheld's methodology with Bain & Company interpretation
• Industry Benchmarking — Compare against BrightLocal 2024 and ReviewTrackers 2024 industry averages
• Platform-Specific Responses — Google (SEO-aware), Amazon (1000-char strict), Yelp (detailed), Etsy (artisan voice)
• FTC 16 CFR Part 255 Compliance — 7-point check on every response ensuring no incentive-for-review-change language
• Brand Voice Injection — Keller's Brand Identity Prism ensures consistent personality across all responses
• Revenue Impact Estimates — Per HBR research, quantifies potential impact of review improvements
• Review Quality Signals — Word count, specificity scoring, verified purchase detection, actionable feedback identification

Feature Comparison

Feature	pf-review-responses (Free)	Podium ($399-599/mo)	Birdeye ($299+/mo)
Response drafting	✅ SERVICE/LAST models	✅ AI suggestions	✅ Smart Reply
Sentiment analysis	✅ ABSA per-topic	Basic positive/negative	Basic positive/negative
Emotion classification	✅ Plutchik 8-emotion	❌	❌
Escalation triage	✅ HEARD 4-level	Basic flagging	Basic flagging
NPS estimation	✅ Reichheld methodology	✅ Survey-based	✅ Survey-based
Industry benchmarks	✅ BrightLocal/ReviewTrackers	❌	✅ Limited
FTC compliance checks	✅ 7-point gate	❌	❌
Platform-specific rules	✅ 6 platforms	✅ Multi-platform	✅ 200+ sites
Live monitoring	❌ Manual export	✅ Real-time	✅ Real-time
Review solicitation	❌	✅ SMS/email	✅ Email
Data privacy	✅ 100% local	Cloud-based	Cloud-based
Price	$0	$4,788-7,188/yr	$3,588+/yr

Estimated Cost per Use

Disclaimer: Token estimates are approximate and based on typical usage patterns measured from skill prompt sizes. Actual costs vary with input data size, conversation length, and complexity. Estimates use Claude Sonnet 4.6 pricing ($3/1M input, $15/1M output). Cowork and Claude Desktop subscription users (Pro/Max/Team) are not charged per-token — these estimates apply only to direct Anthropic API usage. Running stages individually in fresh sessions uses fewer input tokens than running the full pipeline sequentially, because pipeline mode accumulates conversation history across stages.

Per skill (run individually in a fresh session):

Stage	Skill Prompt	User Input	Total Input	Output	Est. Cost
review-analyze	~1.8K	~800	~5.5K	~6.0K	~$0.11
review-respond	~3.0K	~800	~6.6K	~6.0K	~$0.11
review-report	~3.6K	~800	~7.2K	~6.0K	~$0.11
review-load	~4.1K	~2.0K	~8.8K	~2.0K	~$0.06
Standalone total			~28.1K	~20.0K	~$0.38

Full pipeline (all stages in one session — context accumulates):

Stage	Base Input	+ History	Total Input	Output	Est. Cost
review-analyze	~5.6K	0	~5.6K	~6.0K	~$0.11
review-respond	~6.8K	~6.8K	~13.6K	~6.0K	~$0.13
review-report	~7.4K	~13.6K	~21.0K	~6.0K	~$0.15
review-load	~9.1K	~20.4K	~29.5K	~2.0K	~$0.12
Pipeline total			~69.7K	~20.0K	~$0.51

Running the full pipeline once typically costs $0.36–$0.66 in API tokens (Claude Sonnet 4.6).

Known Limitations

• No live platform monitoring — Reviews must be manually exported or pasted. The plugin does not connect to Google, Amazon, Yelp, or other platforms in real-time.
• No review solicitation — Cannot send review request emails or texts to customers (use your platform's native tools for that).
• No multi-user collaboration — One person runs the workflow at a time. Share the output DOCX via email or cloud storage for team review.
• Response drafts require review — AI drafts are starting points, not final copies. Always review and personalize before posting — especially for negative reviews.
• Character limits are informational — Platform character limits (Google: 4,096 chars, Amazon: 1,000 chars) are shown per response but not auto-enforced.
• No auto-posting — Responses must be copy-pasted into each platform manually. This is intentional — human review before posting is a best practice.
• Escalation detection is conservative — The HEARD protocol errs on the side of caution; Level 3+ flags may include reviews that don't require legal review.
• NPS is estimated — Derived from star ratings, not traditional customer surveys. Noted in all outputs per Reichheld methodology limitations.
• Benchmarks are industry averages — Individual business performance varies; top performers significantly exceed these norms.

File Structure

$WORKSPACE/
├── inbound/          # Drop CSV/XLSX review exports here
├── processing/       # Review batches and response JSON (auto-generated)
├── outbound/         # Final DOCX outputs and analysis reports
├── archive/          # Old batch files after archiving
└── .review/
    ├── config.json   # Brand profile, industry benchmarks, framework settings
    └── project.json  # Pipeline state

Important Disclaimers

• AI-Generated Content: This plugin uses AI (LLM) technology which can produce inaccurate or incomplete outputs. All content should be treated as a starting point and reviewed for accuracy before use.
• Not Professional Advice: Outputs do not constitute legal, financial, tax, medical, or other professional advice. Consult qualified professionals before making decisions based on generated content.
• No Compliance Guarantee: References to FTC 16 CFR Part 255, industry standards, and platform policies are for informational purposes only. This plugin does not guarantee compliance with any law or regulation. Users are responsible for verifying all outputs meet their specific regulatory requirements.
• No Endorsement or Affiliation: Mention of BrightLocal, ReviewTrackers, Spiegel Research Center, Harvard Business Review, Podium, Birdeye, or other third-party products and research does not imply endorsement, partnership, or certification by those entities.