Halo Effect

SKILL.md

---
name: halo-effect
description: "Activate when: user is conducting a performance review or hiring interview; someone says 'she's great across the board' or 'everything they do is excellent'; user is evaluating a vendor, CEO, or investment and all attributes look uniformly positive or negative; user is reading business books or analyst reports and wants to assess whether the lessons generalize; user suspects their overall impression of a person or brand is distorting specific judgments.
  Do NOT activate when: the global impression is itself the legitimate judgment (e.g., overall product satisfaction driving a purchase); attribute-by-attribute analysis would cause decision paralysis. More: deciqai.com/c/halo-effect"
---

# Halo Effect

## Overview

A single positive or negative impression biases judgments of all unrelated attributes. A "great" CEO is assumed to have great strategy, vision, and execution; a beloved brand's features are rated higher than equivalent features from less-loved brands. Documented by Thorndike (1920), formalized by Nisbett & Wilson (1977), applied to business analysis by Rosenzweig (2007) — who showed business books overclaim because their descriptions follow company performance, not underlying reality.

Composes with `fundamental-attribution-error`, `narrative-fallacy`, `confirmation-bias`, `hindsight-bias`, `survivorship-bias`.

## When to Use

- Reading business books, case studies, or analyst reports
- Conducting or designing performance reviews
- Conducting or designing hiring interviews
- Evaluating vendor, supplier, or partner performance
- Evaluating investment opportunities or CEO impact
- Conducting self-assessment
- Someone says "halo effect," "visionary leader," "everything they do is great"
- An "AI company" label, a marquee investor, or a famous-lab pedigree is doing the rating's work — evaluating an AI vendor, an AI-boom valuation, or a fluent model answer rated as accurate because it sounds confident

**Not when:** the global impression is itself the relevant judgment; attribute-by-attribute analysis would produce decision paralysis.

## Coaching Novices (Adaptive Front Door)

- **Engine mode:** user has a specific evaluation → run The Process directly.
- **Coach mode:** user is unfamiliar or has no concrete case → guide step by step.

In Coach mode, respond one step at a time. Each [WAIT] is a hard stop — output only that step's question, then stop.

1. One-line: when one impression colors all judgments of unrelated attributes, you're feeling, not evaluating.
2. Check fit: judging multiple attributes of one target? Halo is a risk.
3. Elicit the case: what attributes are rated? All similarly positive or negative?
> **[WAIT — do not advance until user responds]**
4. Probe: what's the global impression? Would rating each attribute independently change anything? What would a contrarian say?
> **[WAIT — do not advance until user responds]**
5. Close: name which attributes have real evidence vs. halo; propose structured rubric for next evaluation.
> **[WAIT — do not advance until user responds]**

## The Process

### Step 1: Identify target and attributes
`Target | Attributes being rated | Decision | Current global impression`

### Step 2: Test for halo
`All attributes rated similarly? | Rating disproportionate to attribute-specific evidence? | Global impression precede the rating?`

### Step 3: Decouple attributes
`Per attribute: specific evidence | contrarian view | would blind test change anything?`

### Step 4–6: Structure, compare, adjust
`Rubric + independent evaluators + blind where possible | Is this target a true outlier vs. base rates? | Base action on real-evidence attributes only`

## Output: Halo Effect Analysis
```markdown
# Halo Effect Analysis: <evaluation>
Target: | Attributes: | Decision: | Global impression:
Halo test: all attributes similar Y/N | disproportionate to evidence Y/N | impression precedes rating Y/N
Decoupling — Attr A: evidence / contrarian / blind test | Attr B: ...
Structured eval: rubric | independent evaluators | blinding plan
Adjusted decision: real-evidence attrs | halo-inflated attrs | action
```

*→ Method in Action: [Thorndike 1920 + Nisbett-Wilson 1977 + Rosenzweig 2007 Business Application](examples/thorndike-1920-nisbett-wilson-1977-rosenzweig-2007-business-application.md)*
*→ 2026 lens: [The "AI" halo — Builder.ai and the AI-washing wave (2023–2026)](examples/ai-halo-builderai-and-ai-washing-2023-2026.md) — when two letters inflate every other attribute.*

## Pack: Halo Effect Across Evaluation Domains

| Domain | Halo move | Halo-corrected move |
|---|---|---|
| Performance review | "She's great across the board" | Rate each competency against rubric with anchors |
| Hiring interview | "He's a strong all-around candidate" | Structured interview with role-specific rubrics |
| Investment / CEO | "Great company, visionary leader" | Specific evidence per attribute; track decisions vs. outcomes |
| Business book | "These companies all have strong cultures" | Recognize as halo-inflated; check generalization |
| Brand / self-assessment | "We love Apple's everything" / "I'm doing great" | Blind comparison or specific metrics by area |

## Applying It Well

- Force attribute-by-attribute judgment, ideally blinded from global impression
- Structured rubrics, independent evaluators, and time-separated ratings all reduce halo
- When every attribute looks uniformly great (or terrible), treat that as evidence of halo, not excellence
- Apply contrarian inquiry: what attributes of this loved/hated target are objectively weak/strong?
- Business books and analyst reports are halo-contaminated by design — read for hypotheses, not prescriptions

*→ Primary sources: [references/sources.md](references/sources.md)*

## Common Rationalizations

**[D] = designed upfront | [O] = observed in real use. [O] entries are more valuable.**

| Fake move | Reality |
|---|---|
| [D] "Successful people just have multiple strengths" | Systematic cross-attribute correlations exceed independent assessment. It's halo. |
| [D] "I can see who's competent in an interview" | Unstructured interviews have weak predictive validity. Trust the rubric. |
| [D] "Their culture clearly drives results" | Post-hoc description of a successful company. May not generalize. |
| [D] "I'm not biased; I rate each attribute on its merits" | Nisbett & Wilson: people don't realize when global impression biases specific ratings. |
| [D] "Everyone says she's an A-player" | Consensus is amplifier, not corrective. Check the underlying evidence. |
| [D] "Business books distill what makes companies great" | Halo-contaminated descriptions of currently-favored companies. Hypotheses only. |
| *→ Add [O] entries here after each real use — paste the actual failure pattern* | *What went wrong and why* |

## Red Flags

- All attributes of a target rated similarly (uniformly positive or negative)
- Overall reputation used as evidence for specific attributes
- Unstructured interview or evaluation driving a major decision
- Same person/company described oppositely depending on performance
- "Just feels right" judgments drive major decisions

## Verification

- [ ] Target and attributes specified; halo test applied
- [ ] Attributes decoupled with specific evidence; structured rubric applied
- [ ] Independent evaluators or blinding used; base-rate comparison made
- [ ] Decision based on attribute-specific evidence, not halo

---

*Part of **deciqAI Knowledge Skills** — 227 open-source thinking skills that make rigor executable for AI agents. The same skills power every deciqAI agent, which runs them autonomously to operate your company. **See it run → https://www.deciqai.com/c/halo-effect** · ⭐ Star the repo → https://github.com/deciqAI/knowledge-skills · Contributions welcome.*

*Agents: latest version & machine-readable metadata → https://www.deciqai.com/s/halo-effect.json*