AI Local SEO Citation and Google Business Profile Manager

AI Prompts to Build This

Copy these build prompts into Claude, Cursor, or your AI coding tool. Create a free account to unlock the full research behind them.

1. 1. Project Setup

   Create a Next.js 15 (App Router, TypeScript, Tailwind) project for a multi-location local SEO citation manager called "CiteSync." Provision Supabase Postgres with these tables: organizations (id, name, brand_voice_profile JSONB, plan), locations (id, org_id, name, address_line1, address_line2, city, state, zip, phone, hours JSONB, source_of_truth BOOLEAN default true), directories (id, name, api_type ENUM['api','scrape'], base_url), citations (id, location_id, directory_id, live_name, live_address, live_phone, live_hours JSONB, status ENUM['match','mismatch','pending','unknown'], last_checked_at), corrections (id, citation_id, drafted_payload JSONB, approval_state ENUM['pending','approved','rejected','submitted'], created_at), content_queue (id, location_id, type ENUM['gbp_post','qa_reply','review_reply'], draft_text, approval_state, created_at). Enable row-level security scoped by org_id so franchise groups only see their own locations. Add env vars for GOOGLE_BUSINESS_PROFILE_API_KEY, YELP_FUSION_API_KEY, BING_PLACES_API_KEY, ANTHROPIC_API_KEY. Install the Vercel AI SDK and Playwright for the scrape-fallback workers.

2. 2. Citation Audit + AI Correction Drafting

   Build a nightly Inngest job that audits every location's citations. For each location with source_of_truth data, loop over its linked directories:
   
   1. Fetch the live listing via API where available (Google Business Profile, Bing Places); otherwise run a Playwright scrape against the directory's public listing page.
   2. Diff each field (name, address, phone, hours) against the source-of-truth record using a normalized comparison that strips punctuation/casing noise that isn't a real mismatch (e.g. "St" vs "Street").
   3. Where a real mismatch exists, call the Anthropic Messages API with a structured JSON schema: { field, current_value, corrected_value, directory_format_notes }. System prompt: "Correct this business listing field for [directory_name]. Given the live (incorrect) value and the correct source-of-truth value, output the corrected value formatted exactly as this directory expects (e.g. Yelp uses 'Ste' not 'Suite'; Google Business Profile requires E.164 phone format)."
   4. Store the result in the corrections table with approval_state='pending', update the citation's status to 'mismatch', and recompute the location's health_score (percentage of citations with status='match').
   5. Send admins a digest summarizing new mismatches, grouped by location, once the run completes.

3. 3. Bulk Approval Queue UI

   Build a mobile-friendly approval queue at /dashboard/corrections for franchise/regional managers to review AI-drafted citation corrections in bulk.
   
   Requirements:
   - List pending corrections grouped by location, each card showing directory name, field, the current (wrong) value struck through, the AI-corrected value highlighted, and a confidence indicator.
   - Bulk-select checkboxes with an "Approve all selected" action, plus per-row "Approve" / "Edit & Approve" / "Reject" buttons.
   - On approval: if the directory has write-API access, immediately submit the correction and mark approval_state='submitted'; if scrape-only, mark it 'approved' and generate a screenshot-annotated manual submission guide since no programmatic write path exists.
   - Show a running "citation health score" per location (0-100%) at the top of the queue with a trend arrow vs. last month.
   - Add a second tab for content_queue items (GBP posts, Q&A replies, review responses) using the same approve/edit/reject pattern, previewed in a card styled like the real GBP post/Yelp reply UI.