UNPKG

major-ai-skills

Version:

Installable agentic skills / AI agent skills (SKILL.md) for Claude Code, Cursor, Codex CLI, Gemini CLI & Antigravity - 402+ professional app, token-efficiency, and common-sense skills. SEO/GEO ready.

104 lines (78 loc) 5.2 kB
--- name: verify-names-and-locations description: "Check names, titles, addresses, and locations against reliable sources before using them in a deliverable." category: common-sense risk: safe source: self source_type: self date_added: "2026-08-26" tags: ["entity-grounding", "named-entities", "hallucination-prevention", "fact-checking", "verification", "prompt-engineering"] tools: ["claude", "cursor", "gemini", "codex", "chatgpt"] --- # Verify Names, Locations, & Entities (Entity Grounding) (AI Skill) ## Overview Large Language Models do not possess a live GPS or relational database of human beings; they generate names and locations based on **statistical token probability**. When asked to generate market research (*"Who is the Head of Procurement at Company X?"* or *"List 5 boutique coffee roasters in Austin, Texas"*), AI models frequently synthesize plausible-sounding names, fake street addresses, and nonexistent executives. The **Entity Grounding Protocol** establishes strict rules for detecting entity hallucinations and verifying proper nouns against external ground truth. --- ## Statistical Plausibility vs. Ground-Truth Reality ``` ┌─────────────────────────────────────────────────────────────┐ Entity Hallucination Trap User asks: "Who is the Chief AI Officer at Acme Corp?" [ LLM TOKENS GENERATE STATISTICAL PLAUSIBILITY ]: "Dr. Sarah Jenkins, former MIT AI Lab researcher..." Plausible sounding, perfectly formatted, 100% FICTION [ THE 4-POINT GROUNDING LADDER (External Verification) ]: 1. Check LinkedIn / Company 'About' Page 2. Verify Secretary of State Corporate Filings 3. Verify Street Address on Google Maps / Postal DB 4. Search Official Press Releases for Appointment Date └─────────────────────────────────────────────────────────────┘ ``` --- ## The 4 High-Risk Entity Danger Zones 1. **Executive & Staff Names**: Corporate turnover is rapid; models hallucinate past or fictional personnel. 2. **Physical Street Addresses & Suite Numbers**: Models generate realistic-looking street numbers that lead to empty lots. 3. **Local Business Names & Phone Numbers**: Phone numbers and store hours change frequently. 4. **Legal & Court Case Citations**: Precedent case names (e.g. *Smith v. Johnson Corp*) are notoriously confabulated. --- ## Master Entity Verification Prompt Templates ### Pattern 1: The Named Entity Sourcing Directive Use when conducting market intelligence, B2B lead generation, or competitive research: ```markdown List the key leadership team at [COMPANY]. Strict Entity Grounding Rules: 1. For every individual listed, provide their **exact job title** and the source URL where this is verified. 2. If their current role cannot be confirmed via your live search tool, mark as: `[STATUS UNVERIFIED / REQUIRES MANUAL CHECK]`. 3. Do NOT extrapolate or guess executive names based on past news. ``` --- ### Pattern 2: The Physical Location & Address Guardrail Use for event planning, logistics, or real estate analysis: ```markdown Suggest 3 conference venues in [CITY] that can host 200 attendees. - State the exact venue name, neighborhood, and physical street address. - Flag any venue that you are unsure is currently in business. ``` --- ## Real-World Case Study ### Scenario: Preparing a B2B Sales Outreach List #### The AI Hallucination Failure > An SDR prompted AI: *"List the VP of Security at 5 mid-market fintech companies."* > - The AI generated 5 names. > - The SDR sent personalized cold emails. > - **3 of the 5 people never worked at those companies**; 1 had left 3 years prior. > - Result: High bounce rate and damaged sender domain reputation. #### The Entity Grounded Workflow > The SDR added the Grounding Rule: *"List the companies and their current publicly stated Security hiring initiatives; provide LinkedIn search query links rather than generating specific personal names."* > - The SDR clicked the 5 verified LinkedIn search links in 60 seconds, identified the exact active VPs, and sent emails with 100% accuracy. --- ## Summary Best Practices - **Never send an email to an AI-generated person's name without checking LinkedIn**: Takes 10 seconds and prevents major embarrassment. - **Search address coordinates**: Always paste addresses into Google Maps before booking travel or printing collateral.