# Beware Of AI Recommendation Bias

> How to neutralize pre-training data bias, SEO affiliate spam influence, and brand popularity bias when evaluating software, hardware, and vendors.

- Skill: `alivirgo/beware-of-ai-recommendation-bias` (Agent Skill)
- Install (CLI): `npx skillmds@latest add alivirgo/beware-of-ai-recommendation-bias`
- Raw SKILL.md: https://api.skillmd.com/api/skills/alivirgo/beware-of-ai-recommendation-bias/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: alivirgo (https://skillmd.com/u/alivirgo)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/alivirgo/beware-of-ai-recommendation-bias

---


# Beware of AI Recommendation & Brand Bias (AI Skill)

## Overview
When you ask an AI *"What is the best CRM software?"* or *"What is the best project management tool?"*, the model does not run an objective laboratory test. Instead, it reflects the **statistical volume of internet training data** - which is overwhelmingly dominated by massive venture-backed marketing budgets, affiliate blog spam, and high-frequency brand mentions.

This skill teaches the **Neutral Vendor Evaluation Protocol**: prompting techniques that neutralize popularity bias, blind-test feature sets, and extract objective trade-offs.

---

## The Sources of AI Recommendation Bias

```
┌─────────────────────────────────────────────────────────────┐
│                 Why AI Favors Certain Brands                │
│                                                             │
│  1. Pre-Training Volume ──► Brands with 10M blog posts dominate│
│  2. Affiliate SEO Spam  ──► "Top 10" listicles bias weights │
│  3. Recency Lag         ──► Newer, superior tools omitted   │
│  4. Sycophancy          ──► Agrees with user's named brand  │
└─────────────────────────────────────────────────────────────┘
```

---

## Master Neutral Evaluation Prompts

### Pattern 1: The Blind Feature-First RFP Audit
Describe your technical and budgetary requirements *without* naming specific brands:

```markdown
Here are my exact requirements for a [SOFTWARE CATEGORY]:
- Budget: [MAX BUDGET / SEAT]
- Must-Have Features: [LIST 3-4 CRITICAL CAPABILITIES]
- Technical Constraints: [SELF-HOSTED / CLOUD / API ACCESS]

Task:
1. Identify 3 competing products that strictly fit these criteria (include at least 1 lesser-known or open-source alternative).
2. For each product, list its #1 fatal limitation or hidden pricing trap.
3. Provide a side-by-side comparison table without marketing buzzwords.
```

---

### Pattern 2: The Symmetrical Trade-Off Engine
Forces equal critical scrutiny across competing tools:

```markdown
Compare [PRODUCT A] vs. [PRODUCT B] vs. [PRODUCT C] for [SPECIFIC TEAM SIZE / USE CASE].

Rules:
- Give equal word count to each product.
- For every product, you MUST provide exactly 2 distinct advantages and 2 severe disadvantages.
- Identify the specific scenario where [PRODUCT A] is the WRONG choice.
- Identify the specific scenario where [PRODUCT B] is the WRONG choice.
```

---

## Real-World Case Study

### Scenario: Choosing a Team Knowledge Base

#### Biased Direct Prompt
> **User**: *"Why should we use Notion for our company wiki?"*
>
> ❌ *Result: AI writes an enthusiastic sales pitch praising Notion's flexibility, completely omitting its lack of enterprise search indexing, slow mobile performance on large databases, and chaotic permission structures.*

#### Neutralized Symmetrical Prompt & Output
> **User**: *"Compare Notion, Confluence, and Slite for an 80-person engineering team. Highlight where each tool fails."*

**AI Response**:

| Tool | Core Advantage | Where It Fails / Fatal Flaw | Ideal Team Match |
| :--- | :--- | :--- | :--- |
| **Notion** | Extreme UI flexibility, custom database views, and rich media blocks. | **Permission chaos & slow search**: Becomes unruly with 50+ users; full-text search across large workspaces is sluggish. | Design & product-led startups under 30 people. |
| **Confluence** | Deep Jira integration, audit compliance, robust permissions. | **Clunky, legacy UX**: Slow load times, complex editor; engineering resistance to updating documentation is high. | Large enterprise orgs tied to Jira/Bitbucket. |
| **Slite / Basecamp** | Blazing fast search, clean markdown-first editor, zero setup. | **Limited database schemas**: Lacks complex relational database and formula capabilities. | Async remote teams who prioritize fast reading/writing over relational databases. |

---

## Critical Rules to Prevent AI Bias
1. **Never ask "Is [Brand X] good?"**: Ask *"What are the 3 biggest operational risks of choosing [Brand X]?"*
2. **Always request an open-source or indie alternative**: Breaks the dominance of venture-funded SaaS tools in the context window.
3. **Inquire about total cost of ownership (TCO)**: Ask *"What are the hidden add-on costs (API limits, seat minimums, SSO paywalls) for each?"*