英语 |[Русский ]

通过心理学而非操控来实现的说服。

一个AI提示技能，能将20种认知偏误转化为高转化率的营销文案——适用于任何语言、任何平台、任何受众。**v2.2：自成一体的单文件，具有叙事深度。**所有表格、13个带有检测规则的反模式、文化数据和少数样本都内嵌在 SKILL.md 中。命名人物故事弧（NPSA）要求确保输出符合人类水平，同时进行机械验证。核心操作无需外部文件读取。

---

这是什么？

MindFluence 是一个针对大型语言模型（LLM）的系统提示（技能），包括GPT、Claude、Gemini、DeepSeek或任何有能力的模型。它将人工智能转变为一个营销策略师，理解为什么人类点击、阅读和购买——而不仅仅是写什么。

它建立在数十年行为经济学和进化心理学研究基础上：卡尼曼的两种体系、恰尔迪尼的原则、芒格的心理误判、费斯廷格的认知失调——被浓缩成可操作的文案写作指导。

它是如何运作的

1. **将SKILL.md**作为系统提示词加载到任意LLM中。仅此而已——文件完全自成一体。

MindFluence

[ English | Русский ]

Persuasion engineered through psychology, not manipulation.

An AI prompt-skill that turns 20 cognitive biases into high-converting marketing copy - for any language, any platform, any audience. v2.3: self-contained core with narrative depth and provenance. All tables, 13 anti-patterns with detection rules, cultural data, and few-shot examples are inlined into SKILL.md. A mandatory PROVENANCE / TRUTH GATE prevents invented numbers, names, and sources. Narrative depth supports two compliant modes: real-data Named Person Story Arc (NPSA) or a labeled [illustrative] fallback. No external file reads required for core operation.

---

What is this?

MindFluence is a system prompt (skill) for LLMs - GPT, Claude, Gemini, DeepSeek or any capable model. It transforms the AI into a marketing strategist who understands why humans click, read, and buy - not just what to write.

It’s built on decades of research in behavioral economics and evolutionary psychology: Kahneman’s two systems, Cialdini’s principles, Munger’s psychological misjudgments, Festinger’s cognitive dissonance - distilled into actionable copywriting instructions.

How it works

1. Load SKILL.md as a system prompt into any LLM. That’s it - the file is self-contained for core operation.
2. Give a task: “Write a LinkedIn post about…” / “Create a landing page for…” / “Audit this ad for cognitive biases.”
3. Get output with every bias labeled, every decision explained, every word psychologically justified.

The LLM uses a single-table Bias Selection Router (audience × product × platform → instant bias stack) instead of a 10-step procedure. After generating, it runs a mechanical Post-Generation Verification - 7 checks (numbers, names, fear exit, scarcity explanation, system-blame, human narrative, provenance) - and a mandatory narrative depth requirement: real-data NPSA or a labeled [illustrative] fallback.

Four modes:

• Quick Mode (default) - data provenance audit → router lookup → scenario override → category check → compose → 7-point verify. Fast for user, fast for LLM.
• Deep Mode - the AI asks 5 clarifying questions at once, then routes through the same lookup.
• Audit Mode - analyzes existing copy for biases and anti-patterns using mechanical detection rules (13 anti-patterns including Statistical-Only Fallacy).
• Optimize Mode - iterates on underperforming copy using funnel-to-bias mapping. Generates A/B variants with statistical confidence thresholds. Also handles qualitative rewrites without metrics: tone shifts, bias injection/removal, anti-pattern fixes, audience/platform adaptation (7 rewrite intents).