How to build auto-publish gating for AI review-responses

The 4-skill bundle on the auto-publish-gating agent

One agent. Four coordinated skills. The Classify + Threshold + Route + Audit bundle runs above the review- management + owner-response-API + LLM-draft substrate and writes one canonical per-review per-location per-draft auto-publish-decision record.

Compliance overlay

Five anchors run per-review per-location per-draft before any owner-response auto-publishes. The first anchor is operationally distinctive: per-confidence-band threshold matrix + per-platform review-response policy + per-vertical professional-licensing-board response + per-franchisor LAM approval + per-confidence calibration converge on every auto-publish decision.

FAQ

What is auto-publish gating for AI review-responses — and what is the per-confidence-band-threshold-times-per-platform-review-response-policy-times-per-vertical-professional-licensing-times-per-franchisor-LAM-approval-times-Article-22-automated-decision-making problem distinctive to this skill?

A franchise multi-store operator with 80-300 locations ships 5,000-50,000 reviews per month across 10-15 platforms (Google Business Profile + Yelp + Apple Maps + Facebook + TripAdvisor + Booking.com + Hotels.com + Expedia + Better Business Bureau + Trustpilot + Glassdoor + Indeed). Each LLM-drafted owner-response (OpenAI + Anthropic + Google Gemini + Mistral + Cohere) must route to auto-publish vs light-review vs full-human-review vs publish-block based on per-confidence-band threshold matrix + per-review-class risk + per-review-sentiment severity + per-reviewer-history + per-platform owner-response policy + per-vertical professional-licensing-board response rules + per-franchisor brand-standards + per-franchise LAM approval. The four-skill bundle on the auto-publish-gating agent — Classify, Threshold, Route, Audit — sits above the review-management + owner-response-API + LLM-draft substrate (Birdeye + Reputation.com + Podium + GatherUp + Grade.us + ReviewTrackers + Yotpo + Trustpilot + ReviewInc + Customer Lobby + Womply + NiceJob + Broadly + Swell + Reviewshake) and writes a per-review per-location per-draft canonical auto-publish-decision record. The operationally distinctive anchor: per-confidence-band auto-publish threshold matrix (tier-1 auto-publish + tier-2 light-review + tier-3 full-human-review + tier-4 publish-block) + per-review-class risk classification (P0 P1 P2 P3 P4 severity) + per-review-sentiment severity (NPS-aligned -100 to +100 + Plutchik emotion wheel + per-aspect sentiment) + per-reviewer-history (per-Google-account + per-Yelp + per-TripAdvisor + previous-reviewer-rating + verified-purchase) + per-Google-Business-Profile review-response policy + Google Business Profile prohibited content + Yelp Content Guidelines + Yelp Owner Response Guidelines + Apple Maps Business reviews + Facebook Community Standards reviews + TripAdvisor Owner Response Guidelines + Booking.com Owner Response + Better Business Bureau Response Standards + Trustpilot Response Code of Conduct + Glassdoor Response Guidelines + per-platform owner-response-API rate-limit (Google Business Profile 100 req/sec + Yelp Fusion 5000 req/day + Facebook Graph 200 req/hour) + per-vertical professional-licensing-board response rules (state bar attorney + state medical board + state dental + state pharmacy + state real estate + state insurance + state CPA + state PE + state architect + state veterinarian) + per-franchisor brand-standards manual + per-franchise LAM approval + per-confidence calibration (Brier + ECE + reliability diagram + isotonic + Platt).

Why do Birdeye + Reputation.com + Podium + GatherUp + Yelp Fusion + Google Business Profile API + OpenAI break at franchise-multi-store-multi-platform-auto-publish-gating scale?

Each review-management + owner-response-API + LLM-draft vendor ships per-account flat draft-then-publish primitive at single-platform level. None coordinates per-review per-location per-draft auto-publish-vs-human-review routing decision against per-confidence-band threshold matrix + per-review-class risk classification + per-review-sentiment severity + per-reviewer-history + per-platform owner-response policy + per-vertical professional-licensing-board response. None handles per-vertical state-bar attorney response prohibition + state-medical-board response prohibition + ABA Model Rule 7.1-7.5 + per-state attorney advertising 50-state convergence. None gates against FTC Endorsement Guides + FTC Fake Review Rule + Tennessee ELVIS Act 2024 + per-state biometric + GDPR Article 22 automated decision-making consent when AI-ML auto-publish gating drives publish-block. None enforces SOX 302/404/906 when public-company franchisor material. None writes a per-review per-location per-draft WORM auto-publish-decision audit trail. The four-skill bundle Classify + Threshold + Route + Audit sits above the review-management + owner-response-API + LLM-draft substrate — it does not replace it.

How does Classify + Threshold work?

Classify runs per-review per-location per-draft classification: per-review-class risk classification (P0 P1 P2 P3 P4 severity), per-review-sentiment severity (NPS-aligned -100 to +100 + Plutchik emotion wheel + per-aspect sentiment via OpenAI + Anthropic + Google Gemini + Mistral + Cohere LLM), per-reviewer-history (per-Google-account history + per-Yelp-account history + per-TripAdvisor account + previous-reviewer-rating-history + verified-purchase tag), per-review-trust-score, per-platform owner-response-policy applicability (Google Business Profile + Yelp + Apple Maps + Facebook + TripAdvisor + Booking.com + Hotels.com + Expedia + Better Business Bureau + Trustpilot + Glassdoor), per-vertical professional-licensing-board applicability (state bar + state medical + state dental + state pharmacy + state real estate + state insurance + state CPA + state PE + state architect + state veterinarian). Per-classification per-confidence calibration: Brier score + Expected Calibration Error (ECE) + reliability diagram + isotonic regression + Platt scaling + conformal prediction. Threshold runs per-classification per-confidence-band auto-publish threshold matrix routing: tier-1 auto-publish (confidence >= 0.92 + risk <= P3 + per-platform policy pass + per-vertical licensing pass + per-franchisor brand-standards pass + per-franchise LAM pre-approval) + tier-2 light-review (confidence 0.80-0.92 + per-franchisor reviewer single-pass) + tier-3 full-human-review (confidence 0.50-0.80 + per-franchisor + per-franchisee dual-pass) + tier-4 publish-block (confidence < 0.50 + risk P0/P1 + per-platform policy fail + per-vertical licensing prohibition + per-franchisor brand-standards fail).

What does Route + Audit do?

Route runs per-review per-draft routing through Twilio + Telnyx + Bandwidth SMS escalation + Slack + Microsoft Teams + Discord franchisee escalation channel. Per-route per-draft severity: P0 platform policy violation immediate (per-Google-BP prohibited content + per-Yelp Content Guidelines + per-Apple Maps) + P1 per-vertical professional-licensing-board response prohibition 72-hour (state bar + state medical) + P2 per-franchisor brand-standards drift 7-day + P3 per-franchisee LAM gap 30-day + P4 docs-only. Gate runs 5 anchors per-review per-location per-draft before any owner-response auto-publishes. (1) Per-confidence-band auto-publish threshold matrix + per-review-class risk + per-review-sentiment severity + per-reviewer-history + per-platform owner-response policy + per-vertical professional-licensing-board response rules + per-franchisor brand-standards + per-franchise LAM approval + per-confidence calibration (Brier + ECE + reliability + isotonic + Platt + conformal) + replication-crisis discipline. (2) FTC Endorsement Guides 16 CFR Part 255 + FTC Fake Review Rule 16 CFR Part 465 + FTC Native Advertising .com Disclosures + Section 5 + Pfizer 1972 + MARS + Health Products + CFPB UDAAP + Lanham + USPTO + state UDTPA + Robinson-Patman + FDD Item 12 + 15-state franchise + per-franchisor brand-standards manual. (3) ABA Model Rule 7.1-7.5 + 1.18 + per-state attorney advertising 50-state + HIPAA 45 CFR 164.502/504/514 Safe Harbor when health/wellness review + state medical board + FINRA Rule 2210 + Rule 3110 when financial-services review + SEC Regulation FD + FDA OPDP + DEA + alcohol TABC/CalABC/TTB + cannabis state-board + tobacco + per-state age-restriction + COPPA. (4) Tennessee ELVIS Act 2024 right-of-publicity when AI-generated owner-response voice/likeness + per-state right-of-publicity + EU AI Act Article 50 transparency when AI-generated review-response + Article 13/14/15 + Annex III when AI-ML auto-publish gating routes publish-block + Article 6/27 FRIA + DSA + DMA + GDPR Article 6/7/22/28/30 (Article 22 automated decision-making consent) + LGPD + DPDP + PIPEDA + Quebec Law 25 + CCPA + CPRA + 18-state + per-state biometric (BIPA + CUBI + Washington biometric) when face/voice. (5) WCAG 2.2 AA + ARIA + EAA + ADA Title III + Section 508 + Section 230 + DMCA Section 512 + SOX 302/404/906 when public-company franchisor material + COSO + Exchange Act 13(b)(2) + SEC Reg S-K. Audit writes a per-review per-location per-draft WORM auto-publish-decision record: review snapshot + draft snapshot + classification + per-confidence calibration + per-anchor gate-pass + per-tier route + per-vertical licensing applicability + per-platform policy applicability + AI-ML provenance + EU AI Act FRIA + GDPR Article 22 automated decision-making log + Tennessee ELVIS Act consent log. Retention: 7-year FTC + 7-year IRS + 7-year HIPAA + 7-year state bar + 6-year SEC + 3-year FINRA + 7-year SOX + GDPR Article 30 + EU AI Act Article 12 + SOC 2 CC7/CC8.

What does this skill connect to on the auto-publish-gating agent and across the swarm?

On the auto-publish-gating agent: per-review per-location per-draft auto-publish routing + per-franchisee escalation + per-franchisor LAM approval + per-platform owner-response publish. Across the swarm: per-location AI review-response drafting (#565 UPSTREAM canonical for LLM-draft) + per-location compliant social drafting (#598 same FDD Item 12 + per-state attorney advertising + per-franchisor brand-standards substrate) + per-location visitor-intent enrichment (#602 same per-franchisor + per-franchise LAM substrate) + integration-drift-monitor agent (#562 + #569 + #570) + tiered pre-filter deterministic gates + per-state-overlay-composer (#599 UPSTREAM canonical for ABA Model Rule + per-state attorney advertising + HIPAA + FINRA per-state) + per-vertical compliance overlay + governance-decision-router five-destination routing. Commercial-pillar parent: /review-response-orchestration.

What does the 6-workstream pre-engagement-baseline reporting cycle look like for this skill?

Every two weeks during the Tier 3 Fractional CMO with AI Swarm engagement, six workstreams report against the pre-engagement baseline. Workstream 1: per-portfolio per-review per-location per-draft auto-publish-gating coverage — reviews monitored + locations covered + platforms covered + drafts generated. Workstream 2: Classify per-review per-draft classification flow — per-review-class risk + per-review-sentiment severity + per-reviewer-history + per-platform policy applicability + per-vertical licensing applicability + per-confidence calibration absorbed. Workstream 3: Threshold per-confidence-band routing flow — tier-1 auto-publish + tier-2 light-review + tier-3 full-human-review + tier-4 publish-block routing. Workstream 4: Route per-draft escalation flow — Twilio + Slack + Teams franchisee escalation + per-franchisor LAM approval + per-platform owner-response API publish. Workstream 5: Regulatory-defense audit coverage — per-confidence-band threshold + per-platform review-response policy + per-vertical professional-licensing-board response + per-franchisor LAM + FTC Endorsement + Fake Review + ABA + HIPAA + FINRA + ELVIS Act + EU AI Act Article 50 + Article 22 + SOX. Workstream 6: FBC feedback-loop pattern-learning — per-review per-draft realized-vs-predicted auto-publish-decision + per-platform policy enforcement retrospective + per-vertical licensing enforcement retrospective.

Related reading

• Parent commercial pillar: review response orchestration
• Sibling build-pillar: per-location AI review-response drafting (#565 UPSTREAM canonical for LLM-draft)
• Sibling build-pillar: per-location compliant social drafting (#598 same FDD Item 12 + per-state attorney advertising substrate)
• Sibling build-pillar: per-state overlay configuration (#599 UPSTREAM canonical for ABA Model Rule + per- state attorney advertising + HIPAA + FINRA per-state)
• Fractional CMO with AI Swarm
• AI Readiness Assessment