AI Truthfulness Rules

Practical anti-trendslop instructions for using AI tools in executive, strategy, research, coding, and business decision workflows.

The goal is simple: use AI to expand thinking, not outsource judgment.

Project links

• RUMI: rumi.run
• LinkedIn: Kevin Wang

What this gives you

• A universal anti-trendslop prompt for any AI assistant.
• An installable agent skill for tools that support skill-style workflows.
• Tool-specific instruction templates for major AI products.
• An executive checklist for reviewing AI-generated recommendations.
• A reusable decision-review prompt for high-stakes work.

Why this exists

Harvard Business Review warned that LLMs can produce "trendslop": polished strategic advice that favors buzzy ideas over context-specific reasoning.

This repository turns that warning into reusable operating rules for major AI tools, including ChatGPT, Claude, Gemini, Perplexity, Microsoft Copilot, Codex, Claude Code, Cursor, and internal enterprise AI assistants.

Source: HBR, "Researchers Asked LLMs for Strategic Advice. They Got 'Trendslop' in Return"

Quick start

Choose the setup that matches your tool.

Option 1: Any AI assistant

Use this for ChatGPT, Claude, Gemini, Perplexity, Copilot, or an internal enterprise assistant.

1. Open prompts/universal-anti-trendslop-system-prompt.md.
2. Copy the prompt.
3. Paste it into your tool's custom instructions, project instructions, system prompt, or saved prompt library.
4. Start important strategy or business prompts with: Follow the anti-trendslop rules before answering.

Option 2: Executive decision review

Use this before acting on an AI-generated recommendation.

1. Open prompts/executive-decision-review-prompt.md.
2. Paste in the recommendation you want to test.
3. Review the evidence, assumptions, alternatives, risks, and minimum validation needed.
4. Use checklists/executive-ai-review-checklist.md before final action.

Option 3: Coding agents and agentic tools

Use this for Codex, Claude Code, Cursor, or other project-level AI agents.

1. Open the relevant file in tool-configs/.
2. Copy it into your project instruction file, such as AGENTS.md, CLAUDE.md, Cursor rules, or project-level AI instructions.
3. Keep the rule close to the project so every future agent run loads the same behavior.
4. For tools that support skills, copy skills/anti-trendslop-truthfulness/SKILL.md into your local skills directory.

Who should use this

• Executives using AI for strategy, performance reviews, or investment decisions.
• Consultants and analysts using AI for synthesis and recommendations.
• Product, marketing, ecommerce, and digital teams using AI in daily workflows.
• Developers using coding agents such as Codex, Claude Code, or Cursor.
• Enterprise AI teams defining reusable assistant behavior.

Core rules

• Evidence first.
• Separate facts, assumptions, and judgment.
• Compare options using the same criteria.
• Challenge the favored answer.
• Treat fashionable advice as a hypothesis, not a conclusion.
• Re-check instead of defending when challenged.
• Name risks, gaps, and minimum validation before action.
• Keep final accountability with humans.

Tool-specific templates

Tool or context	File
Any AI assistant	prompts/universal-anti-trendslop-system-prompt.md
Executive decision review	prompts/executive-decision-review-prompt.md
Skill-compatible agents	skills/anti-trendslop-truthfulness/SKILL.md
ChatGPT custom instructions	tool-configs/chatgpt-custom-instructions.md
Claude project instructions	tool-configs/claude-project-instructions.md
Gemini Gems or custom instructions	tool-configs/gemini-instructions.md
Codex / AGENTS.md	tool-configs/codex-agents.md
Claude Code	tool-configs/claude-code.md
Cursor rules	tool-configs/cursor-rules.md

Recommended enterprise usage

For teams, add these rules to:

• AI assistant custom instructions
• project-level AI instructions
• coding agent rules
• analyst and consultant prompt templates
• enterprise AI governance playbooks
• decision memo templates

For high-stakes decisions, do not rely on AI output unless a human owner has verified the evidence, assumptions, alternatives, and risks.

Repository governance

• Contributions are welcome through GitHub Issues and pull requests.
• See CONTRIBUTING.md for contribution guidance.
• See SECURITY.md for responsible disclosure guidance.
• See CHANGELOG.md for release history.

License

MIT License. Use, adapt, and share freely.