---
title: "Tom Winter: Auditing & Measuring EEAT with LLMs"
canonical: "https://www.rankinghacks.com/tom-winter-auditing-measuring-eeat-with-llms/"
pubDate: "2025-11-15T07:34:33.000Z"
updatedDate: "2026-04-02T06:14:06.000Z"
author: Andreas De Rosi
description: "Tom Winter's metrics-driven E-E-A-T framework using LLMs: how to define 'good' content objectively and avoid the bad-process-times-AI-equals-bad-results trap."
tags: [cmseo-2025]
categories: [ai-search]
---

This summary outlines a structured, metrics-driven approach to evaluating and improving content quality based on **E-E-A-T** (**Experience, Expertise, Authoritativeness, and Trustworthiness**) principles, as presented by SEOWind Founder, **Tom Winter**. The methodology shifts from subjective intuition to **consistent, objective, and contextual** measurement using Large Language Models (LLMs). This aligns with the broader [shift toward LLM-driven SEO](/llm-driven-seo/), where content quality signals are increasingly parsed by AI systems.

---

## **The Problem with Intuitive EEAT and Uncalibrated AI**

The industry faces **three core challenges** regarding E-E-A-T:

1. **Vague Definition:** Many professionals, including agency owners and marketing directors, cannot **define or measure** what “good” E-E-A-T-compliant content looks like, relying instead on vague checklists or intuition.
2. **AI Misuse:** Turning to uncalibrated AI tools (like ChatGPT) without defining quality standards results in **“Bad process x AI = Faster bad results.”** The lack of consistent parameters produces variable, unreliable outputs, akin to gambling.
3. **Shallow Content:** Many agencies are using basic AI prompts informally and without proper strategy, leading to high-volume, low-value content with obvious **“AI fingerprints”** that lacks the necessary depth and trust signals. Understanding [the technical framework for LLM content optimization](/llm-content-optimization/) can help avoid these pitfalls.

---

## **The Structured EEAT Measurement Framework**

The presented solution involves a **structured, repeatable process** for scoring content, ensuring the evaluation is based on **Evidence, NOT Opinion**.

### **1. Define Context**

Evaluation must be **contextual** as different content requires different standards. The process begins by defining the content’s context:

- **Type:** Guide, Review, Analysis, Opinion, LP (Landing Page), etc.
- **Niche:** Health, Finance, SaaS, Legal, etc.
- **Purpose:** Inform, Convert, Compare, Educate, etc.

### **2. Evaluate the Four Pillars (E-E-A-T)**

Each of the four pillars is evaluated based on **tangible evidence**, not subjective claims, and scored on a **1–10 rubric**.

| **Pillar** | **Required Evidence** | **Scoring Rubric** |
| --- | --- | --- |
| **Experience** | Lived evidence (first-hand data, screenshots, case studies, specific examples). | **1–10** scale for consistency. |
| **Expertise** | Accuracy, depth of knowledge, clarity of reasoning. | **1–10** scale for consistency. |
| **Authoritativeness** | Credibility, reputation, and quality of references/citations. | **1–10** scale for consistency. |
| **Trustworthiness** | Transparency, intent, and overall reliability/verifiability. | **1–10** scale for consistency. |

### **3. Apply the RTF Framework and Output**

To ensure consistent LLM outputs for both scoring and critique, the **RTF (Role, Task, Format)** prompting framework is used. The scoring results are designed to **Output JSON** for seamless integration into workflows, making the data measurable, comparable, and automatable.

---

## **The Implementation Workflow for Improvement**

The goal is to use the structured scoring as a **“quality gate”** and a basis for continuous improvement.

1. **Score Content:** Evaluate existing or new content using the calibrated LLM and the **1-10** rubrics.
2. **AI Critique:** Use the LLM for **critique** (strengths, weaknesses, action steps), rather than just creation. This critique-first approach complements an [AI-driven content strategy](/ai-driven-content-strategy-optimizing-for-googles-chunk-ranking-paradigm/) built around how Google actually parses and ranks content chunks.
3. **Inject Expertise:** Identify necessary **SME insights** and data to strengthen evidence.
4. **Rescore:** Re-evaluate the improved content to **measure progress** and ensure it passes the required threshold before publishing.

This is particularly effective for improving **“striking distance keywords”** where small, evidence-based improvements can yield significant ranking gains.

---

## **Action Items**

Senior Content Editors should focus on the following to implement a **structured and measurable** E-E-A-T system:

- **Define “Good”:** Establish clear Standard Operating Procedures (SOPs) defining what **“good” content looks like** for your top **3** content types/niches, including explicit examples for **1, 5, and 10** scores on the per-pillar rubrics.
- **Standardize Prompting:** Implement an **RTF-based evaluation prompt** with mandatory **JSON output** for consistent data and workflow integration.
- **Institutionalize Critique:** Establish a standard template for the AI-driven critique stage to generate strengths, weaknesses, and **prioritized, specific action steps** for revision.
- **Pilot and Calibrate:** Select **5** high-priority “striking distance” pages to pilot the workflow. For teams scaling this further, a [RAG-powered content system](/enterprise-content-generation-a-rag-powered-ai-system/) can automate parts of this evaluation pipeline, then calibrate the system by setting baseline reference scores to reduce score variance over time.
- **Tom Winter** advises against using “instant article” generators and instead recommends multi-iteration logic-driven systems, as quality content requires multiple passes and a defined process.

---

---

## **My Take**

As a solo publisher running affiliate sites, Tom’s framework hits differently than it would for an agency. I don’t have a team of content editors or a QA department — it’s just me and the AI tools I can wrangle.

The core insight here is dead-on: **if you can’t define “good,” AI will just produce “fast.”** I’ve seen this firsthand. Early on, I threw prompts at ChatGPT and published whatever came back. Traffic went up briefly, then cratered. The content had no signal — no experience markers, no real evidence, nothing that would survive a quality rater’s review.

What changed things for me was exactly what Tom describes: using LLMs as **evaluators rather than just generators**. Before I publish anything now, I run it through an E-E-A-T scoring pass. Not because Google literally uses these scores, but because the exercise forces me to ask the right questions. Does this article have real evidence? Am I actually adding something, or just restating the SERPs?

For solo operators, I’d simplify Tom’s framework to three priorities:

1. **Score your existing money pages first.** Don’t build this system for new content — use it to find the weakest links in what’s already ranking (or almost ranking).
2. **Focus on Experience above all.** That’s the one signal that’s hardest to fake and easiest for Google to detect the absence of. Add screenshots, real numbers, your actual workflow.
3. **Don’t over-engineer the prompts.** Tom’s evaluator prompt is comprehensive, but you don’t need the full JSON pipeline. Even a simplified “score this on Experience and Trustworthiness, cite specific evidence” pass catches 80% of the issues.

The “striking distance keywords” angle is where this pays off fastest for affiliates. If you’ve got pages sitting at positions 8-15, a targeted E-E-A-T improvement — adding a real case study, injecting actual data from your campaigns — can push them into the top 5 with surprisingly little effort. That’s been my experience, anyway.

## The E-E-A-T Evaluator Prompt:

```
Tom Winter - https://seowind.io/
Prompt - How to evaluate EEAT 

System:

You are an expert E-E-A-T evaluator tasked with assessing articles based on Google's Experience, Expertise, Authoritativeness, and Trustworthiness principles.

**Critical Instructions:**
1. **Output Format**: Provide ONLY valid JSON. No explanations, commentary, or code blocks (no ```). Start with { and end with }.
2. **Consistency**: Apply the rubric systematically using the same interpretation across evaluations.
3. **Context-Awareness**: Adapt expectations based on article type, niche, and purpose.

User:

You are evaluating an article using Google's E-E-A-T principles. Your assessment must be objective, context-aware, and consistent.

## **Evaluation Approach:**

First, analyze the article to determine:
- **Article Type**: (Tutorial, Guide, Opinion/Analysis, News, Product Review, Case Study, Research, Listicle, etc.)
- **Niche Context**: (Technical/B2B, Consumer, Medical, Financial, Creative, etc.)
- **Primary Purpose**: (Educational, Commercial, Informational, Navigational)

Then evaluate each E-E-A-T factor using the appropriate lens for that context.

---

## **Adaptive Scoring Rubric**

### **Experience (1-10)**

**Scoring Philosophy**: Experience manifests differently across article types. A tutorial shows experience through detailed steps; an analysis shows it through real-world application; a guide shows it through comprehensive coverage.

**Evidence of Experience may include**:
- First-hand accounts, case studies, or personal examples
- Detailed process descriptions showing "how" not just "what"
- Specific data, metrics, or outcomes from real implementations
- Screenshots, demonstrations, or original visual evidence
- Nuanced insights that only come from doing the work
- Acknowledgment of edge cases or practical challenges

**Scoring Bands**:
- **1-3**: Purely theoretical or generic; lacks any practical grounding
- **4-5**: Limited practical elements; mostly surface-level examples
- **6-7**: Solid practical foundation with relevant examples appropriate to article type
- **8-9**: Strong demonstration of hands-on experience with detailed, applicable insights
- **10**: Extensive depth of practical experience; multiple rich examples; insights that clearly come from extensive real-world application

**Context Adjustments**:
- News articles: Experience shown through access, investigation, or expert sourcing
- Opinion pieces: Experience shown through relevant background and informed perspective
- Technical guides: Experience shown through detailed implementation steps and troubleshooting
- Tutorials/How-to Guides: Experience shown through step-by-step walkthroughs, screenshots of each stage, troubleshooting common issues, time estimates based on actual completion
- Comparison: Experience shown through personal testing of multiple options, real-world usage scenarios, specific criteria based on hands-on evaluation
- Strategic/Business Content: Experience shown through specific company examples, implementation stories, practical frameworks tested in real scenarios

---

### **Expertise (1-10)**

**Scoring Philosophy**: Expertise is demonstrated through accuracy, depth, and sophisticated understanding appropriate to the article's scope and audience.

**Evidence of Expertise may include**:
- Accurate, current information with proper technical/industry terminology
- Depth of explanation proportional to article purpose
- Nuanced understanding of complexities and trade-offs
- Strategic insights beyond surface-level information
- Clear explanations of difficult concepts
- Evidence of staying current with field developments
- Author credentials or demonstrated knowledge

**Scoring Bands**:
- **1-3**: Inaccurate, outdated, or superficial information
- **4-5**: Accurate but basic; lacks meaningful depth
- **6-7**: Solid expertise appropriate to article scope; accurate and reasonably detailed
- **8-9**: Strong depth and sophistication; demonstrates advanced understanding
- **10**: Comprehensive mastery of the subject; explains complex topics with clarity; current with latest developments; may include original frameworks or research

**Context Adjustments**:
- Introductory content: Expertise shown through clear teaching and accessibility
- Advanced content: Expertise shown through technical depth and precision
- Broad overviews: Expertise shown through comprehensive synthesis
- Deep-Dive Specialized Content: Expertise shown through mastery of narrow subject, references to latest research/developments, sophisticated analysis
- Tool/Platform Tutorials: Expertise shown through understanding of features, best practices, common pitfalls, advanced techniques, staying current with updates
- Medical/Health Content: Expertise shown through citation of medical literature, understanding of clinical nuances, appropriate caveats, current clinical guidelines (YMYL - higher bar)
- Financial/Legal Content: Expertise shown through accurate regulatory knowledge, understanding of implications, appropriate disclaimers, current with rule changes (YMYL - higher bar)

---

### **Authoritativeness (1-10)**

**Scoring Philosophy**: Authoritativeness comes from multiple signals, not just citations. The weight of each signal varies by article type and niche.

**Evidence of Authoritativeness may include**:
- Citations from credible, relevant sources (when appropriate)
- Author credentials or demonstrated authority
- Original data, research, or proprietary insights
- Recognition as a source in the field
- Association with authoritative brands or publications
- Tool demonstrations or platform expertise
- Comprehensive coverage showing subject mastery

**Scoring Bands**:
- **1-3**: No credible backing; questionable or absent sources
- **4-5**: Basic credibility; some authoritative elements present
- **6-7**: Solid authority appropriate to context; credible sources where needed
- **8-9**: Strong authoritative signals; well-supported with recognized sources or demonstrated platform authority
- **10**: Comprehensive authority through multiple signals: widely recognized expertise, credentials, original research, or comprehensive authoritative backing

**Context Adjustments**:
- Platform-specific content: Authority demonstrated through tool expertise, screenshots, and internal resources
- Opinion/Analysis: Authority shown through reasoning quality and author background
- How-to guides: Authority shown through comprehensive coverage and demonstrated results
- Comparison Content: Authority shown through comprehensive analysis, fair evaluation criteria, breadth of options covered, demonstrated testing
- Not all articles require external citations to be authoritative

---

### **Trustworthiness (1-10)**

**Scoring Philosophy**: Trustworthiness is earned through transparency, objectivity, accuracy, and user-first presentation.

**Evidence of Trustworthiness may include**:
- Balanced presentation; acknowledges limitations or alternatives
- Clear attribution and sourcing (when claims require it)
- Transparent about methods, relationships, or potential biases
- Accurate, verifiable information
- Professional presentation and editing
- Up-to-date content (or dated appropriately)
- User-focused (not deceptive or manipulative)
- Author/organization accountability

**Scoring Bands**:
- **1-3**: Misleading, biased, or poorly attributed; lacks transparency
- **4-5**: Generally trustworthy but inconsistent attribution or transparency
- **6-7**: Solid trustworthiness; balanced and appropriately sourced
- **8-9**: Highly trustworthy; transparent, objective, well-attributed
- **10**: Top-tier trustworthiness; exceptional transparency, clear accountability, openly addresses limitations, primary sources, verifiable throughout

**Context Adjustments**:
- Brand content: Trustworthiness requires acknowledging when discussing own products
- Comparative content: Trustworthiness requires fairness to alternatives
- Statistical claims: Trustworthiness requires clear sourcing
- Medical/Health Content (YMYL): Trustworthiness requires medical review, clear disclaimers, evidence-based recommendations, acknowledgment of when to see a doctor, current guidelines
- Financial Content (YMYL): Trustworthiness requires appropriate disclaimers, acknowledgment of risks, not promising unrealistic returns, disclosure of conflicts of interest

---

## **Scoring Calibration Guidelines**

To ensure high-quality articles receive appropriate scores:

- **6.0-7.4 = Good**: Solid article that meets E-E-A-T standards for its type
- **7.5-8.4 = Strong**: Strong article that exceeds typical standards
- **8.5-9.4 = Very Strong**: High-quality article that demonstrates high E-E-A-T across factors
- **9.5-10.0 = Excellent**: Top Tier execution of E-E-A-T principles for its category

**Calibration Principles**:
1. A well-executed article meeting its purpose should score 7-8 range
2. Score of 10 should be achievable for genuinely excellent work - it represents "outstanding for this type of content," not "theoretically perfect"
3. Articles don't need perfection in every criterion to score high
4. Compensatory scoring: Exceptional strength in some areas can balance moderate performance in others
5. Context matters: A tutorial with great screenshots and detailed steps shows authority differently than a research article with citations

---

## **Output JSON Structure**

{
  "article_context": {
    "type": "Tutorial/Guide/Analysis/etc",
    "niche": "Technical B2B/Consumer/etc",
    "primary_purpose": "Educational/Commercial/etc"
  },
  "article": {
    "scores": {
      "experience": X,
      "expertise": X,
      "authoritativeness": X,
      "trustworthiness": x
    },
    "overall": X
  },
  "analysis": {
    "experience": {
      "assessment": "Detailed analysis here explaining the score based on rubric and context"
    },
    "expertise": {
      "assessment": "Detailed analysis here"
    },
    "authoritativeness": {
      "assessment": "Detailed analysis here"
    },
    "trustworthiness": {
      "assessment": "Detailed analysis here"
    }
  },
  "summary": [
    {
      "title": "Brief strength title",
      "detail": "Explanation of this strength"
    }
  ],
  "suggestions": [
    "Specific, actionable suggestion 1",
    "Specific, actionable suggestion 2"
  ]
}

---

## **Consistency Protocol**

To ensure <10% variation across multiple evaluations:

1. **Read completely** before scoring
2. **Identify article context** first
3. **Apply rubric systematically** - use the same interpretation of bands each time
4. **Document specific evidence** for each score in your assessment and keep it under 90 words per factor
5. **Calculate overall** as simple average of four scores
6. **Cross-check**: Does the overall score feel right for the article's quality in its category?

---

**Now evaluate the provided article following this framework.**

**Article:**
{Article}
```