# FAQ Quality Research

**Last Updated:** 2026-01-15

Research findings on FAQ best practices for SEO/AEO/GEO optimization, topic relevance validation, and semantic similarity thresholds.

## SEO/AEO/GEO Best Practices for FAQ Ordering

### Logical Flow & Ordering

Based on 2026 best practices, FAQs should follow this ordering:

1. **Intro & Purpose Statement** - High-level summary of what the FAQ covers
2. **Top-Level Core Questions First** - Foundational questions (definitions, "what is", "how does", basics)
   - High-frequency and high-impact queries
   - Answer-first approach: direct answer immediately after question
   - Target 40-80 words for optimal AI visibility
3. **Follow-Up or Clarification Questions** - Common misconceptions, alternatives, when to use
4. **Troubleshooting / Edge Cases** - Rare use-cases, exceptions, long-tail traffic
5. **Localized or GEO-Tailored FAQs** (if applicable) - Location-based questions
6. **Schema & Metadata** - Structured data, last updated dates, governance notes

### Content Best Practices

| Element               | Best Practice                                               | Why                                            |
| --------------------- | ----------------------------------------------------------- | ---------------------------------------------- |
| **Question phrasing** | Natural language, full questions, long-tail                 | Matches user intent and AI queries             |
| **Answers**           | Lead with answer in 1-2 sentences, 40-80 words, then expand | Optimizes for snippet/AI overview pullouts     |
| **Format**            | Headings (H2/H3), short paragraphs, lists, tables           | Improves scannability and model extractability |

### Common Mistakes to Avoid

- Answers too long (>80 words) or vague
- FAQ content only in schema (not visible HTML)
- Keyword stuffing in unnatural ways
- Thin or duplicate content across pages

## Topic Relevance Validation

### Semantic Similarity Thresholds

For FAQ topic relevance validation:

- **Question-to-topic matching**: Start with **0.3-0.4** threshold for relevance
- **Question-to-question matching**: Use **0.6-0.7** for identifying related questions
- **Answer matching**: Use **0.8-0.85** for duplicate answer detection

### Factors Affecting Thresholds

1. **Domain specificity**: More specific domains need higher thresholds
2. **Embedding model quality**: Different models produce different similarity distributions
3. **Question vs answer text**: Questions tend to yield lower similarity than full answers
4. **False positives vs false negatives**: Balance based on user tolerance

### Recommended Thresholds

- **Topic relevance**: 0.3 minimum (questions must relate to post topic)
- **Duplicate detection**: 0.7-0.8 for questions, 0.8-0.85 for answers
- **High-risk domains**: 0.9+ for strict matching

## Semantic Similarity for Duplicate Detection

### Common Threshold Ranges

| Use Case              | Typical Threshold | Notes                                           |
| --------------------- | ----------------- | ----------------------------------------------- |
| FAQ question matching | 0.6-0.8           | Balanced generalization without false positives |
| FAQ answer matching   | 0.8+              | Longer content demands stronger match           |
| Closed-domain systems | 0.85-0.95         | When mistakes are costly (legal, medical)       |
| Open-domain systems   | 0.5-0.7           | Broader net, more risk acceptable               |

### Tuning Recommendations

1. **Start conservative**: Begin with 0.8-0.85 for duplicate detection
2. **Use human review**: For borderline cases (0.75-0.85)
3. **Combine signals**: Use similarity + keywords + metadata for accuracy
4. **Monitor per topic**: Thresholds may behave differently across topics
5. **Iterative adjustment**: Deploy conservatively, monitor, adjust

### Trade-Offs

| Risk                | Threshold Too High         | Threshold Too Low     |
| ------------------- | -------------------------- | --------------------- |
| **False Negatives** | Duplicates not detected    | Fewer false negatives |
| **False Positives** | Rare; distinct FAQs merged | Many; wrong grouping  |

## Implementation Guidelines

### For Our Blog FAQs

Based on research, we should:

1. **Topic Relevance**: Use 0.3 threshold minimum (questions must relate to post)
2. **Duplicate Detection**: Use 0.7 for questions, 0.8 for answers
3. **Ordering**: Follow logical flow (definition → how-to → details → edge cases)
4. **Answer Length**: Target 40-80 words per answer
5. **Pattern Detection**: Identify and remove nonsensical patterns
6. **Human Review**: Always review borderline cases manually

### Quality Standards

- **Topic relevance**: > 0.3 for all FAQs
- **Duplicate similarity**: < 0.7 for questions, < 0.8 for answers
- **Answer length**: 40-80 words
- **Logical ordering**: Definitions before how-to, high-volume queries first
- **No pattern violations**: Remove nonsensical patterns

## Sources

- AEO/GEO Best Practices: Various SEO and AEO optimization guides (2026)
- Semantic Similarity Thresholds: AWS QnABot documentation, developer guides
- Duplicate Detection: Research papers and industry best practices