Concevez un centre d aide qui réduit les tickets répétés et donne à l IA support des sources sûres. — Claude Skill

Intégration IA

Architecture de chatbot IA, pipelines RAG et intégrations plateforme pour centres d aide.

Contenu

• Architecture moderne de support IA (2025-2026)
• Design de pipeline RAG
• Configuration de recherche sémantique
• Écriture de contenu compatible IA
• IA riche en mémoire (tendance 2026)
• Capacités IA agentiques
• Setup IA par plateforme
• Escalade & handoff
• Monitoring & optimisation

Architecture moderne de support IA (2025-2026)

Flow support AI-first

AI-FIRST SUPPORT FLOW (2025-2026)

User query
  -> Intent classification (question vs task, topic, urgency)
  -> Semantic search (RAG) (embedding, vector search, retrieval)
  -> Response generation (answer, citations/links, confidence score)

If confidence is high: direct answer + sources
If confidence is medium: answer + "Was this helpful?"
If confidence is low: ask a clarifying question or escalate

Types de résolution

Design de pipeline RAG

Stratégie de chunking documentaire

CHUNKING PARAMETERS

Chunk size: 500-1000 tokens (optimal for retrieval)
Overlap: 50-100 tokens (preserve context)
Boundaries: Respect section headers, paragraphs

CHUNKING METHODS

1. Fixed-size: Simple, consistent
2. Semantic: Split by meaning (paragraphs, sections)
3. Hierarchical: Parent-child relationships

RECOMMENDED: Semantic chunking with header preservation

EXAMPLE

Original article (2000 tokens):
- Chunk 1: Title + Intro (400 tokens)
- Chunk 2: Section 1 (500 tokens)
- Chunk 3: Section 2 (500 tokens)
- Chunk 4: Section 3 + Conclusion (600 tokens)

Metadata per chunk:
- article_id
- section_title
- position (1/4, 2/4, etc.)
- url
- last_updated

Choix du modèle d embedding

Recommandation : commencer avec text-embedding-3-small, upgrader si la qualité pose problème.

Options de base vectorielle

Configuration retrieval

RETRIEVAL PARAMETERS

Top-K: 3-5 chunks (balance relevance vs. context)
Similarity threshold: 0.7-0.8 (filter weak matches)
Reranking: Yes (improves precision)

HYBRID SEARCH (Recommended)

Combine:
1. Semantic search (70% weight) - meaning
2. Keyword search (30% weight) - exact matches

Benefits:
- Catches exact error messages
- Handles product names, codes
- Better coverage than semantic alone

Assemblage du contexte

PROMPT TEMPLATE

You are a helpful support assistant for [Product].
Answer the user's question using ONLY the provided context.
If the context doesn't contain the answer, say so.
Always cite your sources.

Context:
---
{retrieved_chunks}
---

User Question: {query}

Instructions:
- Be concise and direct
- Use bullet points for steps
- Include relevant links
- If unsure, offer to connect with human support

Configuration de recherche sémantique

Traitement des requêtes

QUERY ENHANCEMENT

1. Spell correction
   "passowrd reset" -> "password reset"

2. Synonym expansion
   "cost" -> "cost OR pricing OR price"

3. Query rewriting (LLM)
   "it's not working" -> "troubleshooting [detected feature]"

4. Intent extraction
   "how do I..." -> how-to intent
   "why is..." -> troubleshooting intent
   "what is..." -> conceptual intent

Ranking des résultats

RANKING SIGNALS

1. Vector similarity score (0.0-1.0)
2. Keyword match (BM25)
3. Recency boost (newer content)
4. Popularity (view count)
5. Manual boost (featured content)

COMBINED SCORE

final_score = (
    0.5 * semantic_score +
    0.3 * keyword_score +
    0.1 * recency_score +
    0.1 * popularity_score
)

Gestion des edge cases

Écriture compatible IA

Structure pour consommation IA

CONTENT RULES FOR RAG

DO:
- Clear, keyword-rich headings
- One concept per paragraph
- Explicit step numbering
- Tables for structured data
- Exact error messages (searchable)
- FAQ format (question as heading)

DON'T:
- Ambiguous pronouns ("it", "this")
- Implicit assumptions
- Marketing fluff in support docs
- Information buried in paragraphs
- Duplicate content across articles

Métadonnées pour IA

ARTICLE FRONTMATTER

---
title: How to Reset Your Password
description: Step-by-step guide to reset password via email or phone
keywords: [password, reset, forgot, login, access]
category: account/security
audience: all-users
difficulty: beginner
last_updated: 2025-01-15
related: [enable-2fa, account-recovery, login-issues]
---

Optimisation extraction de réponse

STRUCTURE FOR DIRECT ANSWERS

Bad (AI must parse):
"You can find your API key in several places.
One option is the dashboard. Another is the
settings page under API section."

Good (AI extracts easily):
"Find your API key:
1. Go to Settings > API
2. Click 'Reveal Key'
3. Copy the key

Alternative: Dashboard > Quick Actions > API Key"

IA riche en mémoire (tendance 2026)

Contrairement aux chatbots stateless, l IA riche en mémoire conserve le contexte entre sessions pour un support plus rapide et plus personnalisé.

Capacités clés

MEMORY-RICH AI BENEFITS

1. Context Retention
   - Remember previous conversations
   - Track user preferences
   - Recall past issues/resolutions

2. Personalization at Scale
   - Tailored responses based on history
   - Proactive suggestions from patterns
   - Reduced "repeat yourself" frustration

3. Faster Resolution
   - Skip re-identification steps
   - Reference previous context
   - Build on prior interactions

Pattern d implémentation

MEMORY ARCHITECTURE

Session Start:
1. Retrieve user profile from CRM
2. Fetch last 5 conversation summaries from vector DB
3. Load relevant context into system prompt

During Conversation:
4. Store key facts extracted by LLM
5. Update preference signals
6. Track resolution outcomes

Session End:
7. Generate conversation summary
8. Store embeddings for future retrieval
9. Update user profile with new signals

STORAGE OPTIONS

- Short-term: Redis (session data, 24hr TTL)
- Long-term: Vector DB (conversation embeddings)
- Structured: PostgreSQL (user profiles, preferences)

Requête de récupération mémoire

# Example: Retrieve relevant past context
def get_user_memory(user_id: str, current_query: str, limit: int = 5):
    # 1. Get user profile
    profile = db.get_user_profile(user_id)

    # 2. Semantic search past conversations
    query_embedding = embed(current_query)
    past_contexts = vector_db.search(
        collection="conversations",
        filter={"user_id": user_id},
        vector=query_embedding,
        limit=limit
    )

    # 3. Assemble memory context
    return {
        "profile": profile,
        "past_interactions": past_contexts,
        "preferences": profile.get("preferences", {})
    }

Capacités IA agentiques

Exécution de tâches (2025-2026)

AGENTIC ACTIONS

Level 1: Information retrieval
- Search knowledge base
- Summarize articles
- Provide links

Level 2: Simple actions
- Create support ticket
- Check order status
- Look up account info

Level 3: Transactional
- Process refund
- Cancel subscription
- Update account details

Level 4: Complex workflows
- Book appointment
- Escalate with context
- Multi-system lookup

Intégration d outils (Function Calling)

TOOL DEFINITIONS (Example)

tools = [
    {
        "name": "check_order_status",
        "description": "Check the status of a customer order",
        "parameters": {
            "order_id": {"type": "string", "required": True}
        }
    },
    {
        "name": "process_refund",
        "description": "Process a refund for an order",
        "parameters": {
            "order_id": {"type": "string", "required": True},
            "reason": {"type": "string", "required": True},
            "amount": {"type": "number", "required": False}
        }
    },
    {
        "name": "create_ticket",
        "description": "Create a support ticket for human review",
        "parameters": {
            "subject": {"type": "string", "required": True},
            "description": {"type": "string", "required": True},
            "priority": {"type": "string", "enum": ["low", "medium", "high"]}
        }
    }
]

Model Context Protocol (MCP)

MCP INTEGRATION (2025)

Purpose: Standardized protocol for AI-to-tool communication

Benefits:
- Plug-and-play tool connections
- Consistent authentication
- Built-in safety guardrails

Use cases:
- Connect AI to CRM (Salesforce, HubSpot)
- Access order management systems
- Query internal databases
- Trigger workflow automation

Setup IA par plateforme

Zendesk AI

ZENDESK AI FEATURES

1. Answer Bot
   - Suggests articles during ticket creation
   - Auto-resolve common questions
   - Learns from agent responses

2. Generative AI (2024+)
   - Draft article summaries
   - Suggest article updates
   - Tone adjustment

3. Intelligent Triage
   - Auto-categorize tickets
   - Priority prediction
   - Agent routing

SETUP STEPS

1. Enable AI in Admin > AI > Bots
2. Train on knowledge base
3. Set confidence thresholds
4. Configure escalation rules
5. Monitor resolution rates

Intercom Fin AI

FIN AI FEATURES

1. Resolution
   - Answers from your content
   - Multi-turn conversations
   - Task execution (with tools)

2. Sources
   - Help Center articles
   - Website content
   - Custom data sources

3. Behavior
   - Customizable persona
   - Handoff rules
   - Business hours

PRICING

$0.99 per resolution
Resolution = AI successfully answers without human

SETUP STEPS

1. Install Fin (Settings > Fin)
2. Connect content sources
3. Test in Sandbox
4. Set live traffic %
5. Monitor Fin reports

Freshdesk Freddy AI

FREDDY AI FEATURES

1. Auto-suggest
   - Canned responses
   - Solution articles
   - Similar tickets

2. Ticket classification
   - Category prediction
   - Priority assignment
   - Group routing

3. Customer-facing bot
   - Self-service answers
   - Ticket deflection
   - Agent handoff

INCLUDED IN: Pro ($49) and Enterprise plans

SETUP STEPS

1. Admin > Freddy > Enable
2. Train on ticket history
3. Configure bot flows
4. Set escalation triggers
5. Review suggestions quality

Implémentation IA custom

BUILD YOUR OWN (Stack)

Frontend:
- Chat widget (custom or open-source)
- WebSocket for real-time

Backend:
- FastAPI / Node.js
- Message queue (Redis)
- Session management

AI Layer:
- LLM (Claude, GPT-4, Llama)
- RAG pipeline
- Function calling

Vector DB:
- Pinecone / Qdrant / pgvector

Integrations:
- Helpdesk API (tickets)
- CRM API (customer data)
- Webhooks (notifications)

Escalade & handoff

Déclencheurs d escalade

AUTO-ESCALATE WHEN

Confidence-based:
- AI confidence < 0.5
- Multiple failed attempts (>2)
- User frustration detected

Content-based:
- Billing disputes
- Legal/compliance
- Security incidents
- VIP customers

Explicit:
- User requests human
- Keywords: "speak to agent", "manager"

Bonnes pratiques de handoff

SEAMLESS HANDOFF

1. Context transfer
   - Full conversation history
   - AI's attempted answers
   - Detected intent
   - Customer info

2. Warm introduction
   "[Agent name] will continue helping you.
   I've shared our conversation so you won't
   need to repeat anything."

3. No dead ends
   - Always offer alternative if no agents
   - Callback option
   - Email follow-up

Collaboration humain-IA

AGENT ASSIST FEATURES

1. Suggested responses
   - Based on conversation context
   - From knowledge base
   - From similar resolved tickets

2. Real-time guidance
   - Policy reminders
   - Upsell opportunities
   - Compliance warnings

3. Auto-summarization
   - Ticket summary after resolution
   - Key points extraction
   - Follow-up suggestions

Monitoring & optimisation

Métriques de performance IA

Assurance qualité

AI QA PROCESS

Weekly:
- Review 50 random AI conversations
- Check accuracy of answers
- Identify hallucinations
- Flag edge cases

Monthly:
- Update content gaps found
- Retrain on new content
- Adjust confidence thresholds
- Review escalation patterns

Quarterly:
- Full accuracy audit
- Benchmark against competitors
- User satisfaction survey
- Cost-benefit analysis

Amélioration continue

FEEDBACK LOOP

1. Collect signals
   - Thumbs up/down
   - "Was this helpful?"
   - Escalation after AI answer
   - User corrections

2. Analyze patterns
   - Common failure modes
   - Missing content topics
   - Misunderstood queries

3. Improve
   - Add/update content
   - Tune prompts
   - Adjust thresholds
   - Add synonyms

Templates d articles

Templates copier-coller pour tous les types d articles de centre d aide.

Contenu

• Template article How-To
• Template article Troubleshooting
• Template article conceptuel
• Template article FAQ
• Template article référence
• Template script tutoriel vidéo
• Checklist production
• Guidelines de contenu visuel

Template article How-To

# How to [Action Verb] [Object]

[1-2 sentence intro explaining what this guide covers and the outcome]

## Prerequisites

- [Requirement 1 - e.g., Admin access required]
- [Requirement 2 - e.g., Feature enabled in Settings]
- [Requirement 3 - optional, link to setup guide]

## Steps

### Step 1: [Action verb + specific action]

[2-3 sentences explaining what to do]

![Screenshot description](path/to/screenshot.png)
*Caption: What the user should see*

### Step 2: [Action verb + specific action]

[Instructions]

> **Note**: [Important callout if needed]

### Step 3: [Action verb + specific action]

[Instructions]

Code block if relevant

## Result

[Describe what success looks like - what the user should see/experience]

![Success state screenshot](path/to/success.png)

## Troubleshooting

| Issue | Solution |
|-------|----------|
| [Common problem 1] | [Quick fix] |
| [Common problem 2] | [Quick fix or link] |

## Next Steps

- [Related task 1](link)
- [Related task 2](link)
- [Advanced guide](link)

---

**Was this helpful?** [Yes] [No]

*Last updated: YYYY-MM-DD*

Guidelines de rédaction How-To

Template article Troubleshooting

# Fix: [Error Message or Problem Description]

[Brief description of the issue and its impact]

## Symptoms

- [What the user sees - exact error text]
- [Related behavior]
- [When it typically occurs]

**Error Message:**

[Exact error text user sees]

## Quick Fixes

Try these solutions in order:

### 1. [Most common solution]

**Why this works**: [Brief explanation]

**Steps:**
1. [Step 1]
2. [Step 2]
3. [Step 3]

**Expected result**: [What should happen]

---

### 2. [Second most common solution]

**Why this works**: [Brief explanation]

**Steps:**
1. [Step 1]
2. [Step 2]

---

### 3. [Edge case solution]

**When to try**: [Specific condition]

**Steps:**
1. [Step 1]
2. [Step 2]

## Root Causes

| Cause | Likelihood | Solution |
|-------|------------|----------|
| [Cause 1] | Common | Solution 1 above |
| [Cause 2] | Occasional | Solution 2 above |
| [Cause 3] | Rare | Contact support |

## Prevention

- [How to avoid this in the future]
- [Best practice recommendation]

## Still Not Working?

If none of the solutions above resolved your issue:

1. **Gather this information:**
   - Browser/app version
   - Steps to reproduce
   - Screenshot of error

2. **Contact support:**
   [Contact Support](link) — Average response: [X hours]

---

**Was this helpful?** [Yes] [No]

*Last updated: YYYY-MM-DD*

Guidelines de rédaction Troubleshooting

Template article conceptuel

# [Concept Name]: [Brief Description]

[2-3 sentence overview explaining what this is and why it matters]

## What is [Concept]?

[Clear definition in plain language, 2-4 sentences]

### Key Points

- [Essential point 1]
- [Essential point 2]
- [Essential point 3]

## How [Concept] Works

[Explanation with diagram or visual if helpful]

[Simple diagram using ASCII or embedded image]

### Components

| Component | Purpose | Example |
|-----------|---------|---------|
| [Part 1] | [What it does] | [Concrete example] |
| [Part 2] | [What it does] | [Concrete example] |
| [Part 3] | [What it does] | [Concrete example] |

## When to Use [Concept]

**Use when:**
- [Scenario 1]
- [Scenario 2]

**Don't use when:**
- [Anti-pattern 1]
- [Alternative approach]

## Examples

### Example 1: [Common use case]

[Concrete example with before/after or input/output]

### Example 2: [Advanced use case]

[Second example showing more complex application]

## Related Concepts

- **[Related concept 1]**: [How it relates](link)
- **[Related concept 2]**: [How it relates](link)

## Learn More

- [How-to guide using this concept](link)
- [Advanced documentation](link)
- [Video tutorial](link)

---

**Was this helpful?** [Yes] [No]

*Last updated: YYYY-MM-DD*

Template article FAQ

# [Topic] FAQs

Frequently asked questions about [topic].

---

## Getting Started

<details>
<summary><strong>Q: [Question in natural language]?</strong></summary>

[Answer in 2-4 sentences]

[Link to detailed guide if needed](link)

</details>

<details>
<summary><strong>Q: [Question 2]?</strong></summary>

[Answer]

</details>

---

## [Category 2]

<details>
<summary><strong>Q: [Question]?</strong></summary>

[Answer]

| Option | Result |
|--------|--------|
| [A] | [What happens] |
| [B] | [What happens] |

</details>

<details>
<summary><strong>Q: [Question]?</strong></summary>

[Answer]

> **Tip**: [Helpful additional info]

</details>

---

## Billing & Account

<details>
<summary><strong>Q: [Billing question]?</strong></summary>

[Answer]

**Related**: [Billing settings](link)

</details>

---

## Troubleshooting

<details>
<summary><strong>Q: Why am I seeing [error]?</strong></summary>

This usually happens when [cause].

**Quick fix:**
1. [Step 1]
2. [Step 2]

**Still not working?** [Contact support](link)

</details>

---

**Can't find your answer?**

- [Search help center](link)
- [Contact support](link)
- [Community forum](link)

*Last updated: YYYY-MM-DD*

Guidelines de rédaction FAQ

Template article référence

# [Feature/API] Reference

Complete reference for [feature/API name].

## Overview

| Property | Value |
|----------|-------|
| **Availability** | [Plan tier] |
| **API Endpoint** | `[endpoint]` |
| **Rate Limit** | [X requests/minute] |
| **Last Updated** | [Date] |

## Parameters

### Required Parameters

| Parameter | Type | Description |
|-----------|------|-------------|
| `param1` | string | [Description] |
| `param2` | integer | [Description] |

### Optional Parameters

| Parameter | Type | Default | Description |
|-----------|------|---------|-------------|
| `option1` | boolean | `false` | [Description] |
| `option2` | string | `null` | [Description] |

## Examples

### Basic Usage

```json
{
  "param1": "value",
  "param2": 123
}

Response:

{
  "status": "success",
  "data": { ... }
}

Advanced Usage

{
  "param1": "value",
  "param2": 123,
  "option1": true
}

Error Codes

Limits & Quotas

Changelog

Related

• API Authentication
• Webhooks Reference
• SDK Documentation

Last updated: YYYY-MM-DD

## Template script tutoriel vidéo

```markdown
# Video: How to [Action]

**Duration**: [X:XX]
**Skill Level**: [Beginner/Intermediate/Advanced]

## Script

### Intro (0:00-0:15)

"In this video, you'll learn how to [outcome]. By the end, you'll be able to [specific skill]."

### Section 1: [Topic] (0:15-1:00)

**Visuals**: [Screen recording of X]

"First, let's [action]. Navigate to [location]..."

**Key Points to Show**:
- [ ] [Visual element 1]
- [ ] [Visual element 2]

### Section 2: [Topic] (1:00-2:00)

**Visuals**: [Screen recording of Y]

"Now that we've [previous action], let's [next action]..."

### Section 3: [Topic] (2:00-3:00)

**Visuals**: [Result/confirmation screen]

"You've successfully [outcome]. Here's what you should see..."

### Outro (3:00-3:30)

"That's how you [action]. For more help, check the links in the description. If you found this helpful, [CTA]."

## Production Checklist

- [ ] Script approved
- [ ] Screen recording captured
- [ ] Voiceover recorded
- [ ] Captions added
- [ ] Thumbnail created
- [ ] Chapter markers set
- [ ] Description with links
- [ ] Published to: [platforms]

## Metadata

**Title**: How to [Action] | [Product Name]
**Description**: Learn how to [action] in [time]. This tutorial covers [topics]. Timestamps: [chapters]
**Tags**: [tag1], [tag2], [tag3]
**Thumbnail**: [Description]

Checklist qualité contenu

Avant publication

QUALITY GATES

[ ] Title matches search intent
[ ] Intro answers "what will I learn?"
[ ] Steps are numbered and actionable
[ ] Screenshots are current (check version)
[ ] Links work (test all)
[ ] Mobile-friendly formatting
[ ] Accessibility: alt text, captions
[ ] Related articles linked
[ ] Feedback mechanism present
[ ] Last updated date set

AI-FRIENDLY CHECKS

[ ] Clear headings with keywords
[ ] No ambiguous pronouns
[ ] Error messages exact (for search)
[ ] No duplicate content elsewhere
[ ] Structured data (tables, lists)

Planning de revue du contenu

Guidelines de contenu visuel

Captures d écran

SCREENSHOT REQUIREMENTS

Size: 1200x800px minimum (2x for retina)
Format: PNG for UI, GIF for sequences
Annotations:
  - Red boxes for emphasis
  - Numbered callouts for steps
  - Blur sensitive data
File naming: [article-slug]-step-[N].png

Enregistrements GIF

GIF GUIDELINES

Duration: 5-15 seconds
Frame rate: 10-15 fps
Size: Under 5MB
Tools: CleanShot, Kap, LICEcap
Use for: Multi-step actions, hover states

Diagrammes

DIAGRAM TYPES

Flowcharts: Decision processes
Architecture: System overviews
Timelines: Sequences, processes
Comparison: Feature matrices

Tools: Excalidraw, Mermaid, Whimsical
Style: Consistent colors, minimal text

Opérations de connaissance

Gouvernance et cadence opérationnelle pour maintenir dans le temps un centre d aide de haute qualité, prêt pour l IA.

Contenu

• Modèle de gouvernance
• Cycle de vie du contenu
• Signaux de fraîcheur et qualité
• Intégration releases et incidents
• Localisation et accessibilité
• Alignement support IA
• Cadence opérationnelle

Modèle de gouvernance

Définissez un ownership clair afin que le contenu reste correct, actuel et sûr.

Rôles recommandés :

• Owner centre d aide (owner programme, priorisation, standards)
• Support operations (outillage, workflows, reporting)
• SMEs produit (exactitude technique)
• Reviewer juridique/sécurité (si requis)
• Rédacteurs/éditeurs (clarté, cohérence, UX)

Assignez l ownership à deux niveaux :

• Owner de catégorie : responsable de la santé d une zone de taxonomie
• Owner d article top : responsable des articles à plus fort impact dans cette zone

Cycle de vie du contenu

Utilisez un cycle cohérent pour éviter la dérive :

1. Intake
   • Sources : tickets, logs de recherche, escalades, release notes, incidents.
2. Draft
   • Utiliser templates standard et règles d écriture compatibles IA.
3. Review
   • Approbation SME pour exactitude; revue juridique/sécurité si nécessaire.
4. Publish
   • Garantir le bon placement IA, les tags et les liens internes.
5. Measure
   • Suivre utilité, succès de recherche et escalade après lecture.
6. Improve
   • Réécrire titres, ajouter visuels et corriger les prérequis manquants.
7. Retire
   • Rediriger URLs obsolètes; archiver contenu déprécié avec rationale.

Signaux de fraîcheur et qualité

Utilisez à la fois des signaux temporels et comportementaux.

Signaux de fraîcheur :

• Releases produit affectant une fonctionnalité référencée dans l article
• Liens cassés, captures obsolètes ou labels UI changés
• Article non mis à jour depuis 6-12 mois (seuil selon cadence de release)

Signaux comportementaux :

• Taux élevé search-to-exit (les utilisateurs abandonnent après recherche)
• Taux élevé d escalade après vue article (le contenu ne résout pas le problème)
• Taux élevé de feedback négatif (thumbs down, note basse)
• Taux élevé de vues répétées pour le même problème (besoin de plusieurs lectures)

Heuristique de priorisation :

• Corriger le plus petit nombre d articles qui défléchissent le plus grand nombre de tickets.

Intégration release et incident

Faire des updates contenu une partie standard de la livraison :

• Pour chaque release qui change UI/workflows, mettre à jour les articles how-to et troubleshooting impactés.
• Pour chaque incident, publier :
  • article "Statut et contournement" (pendant l incident)
  • explication post-incident et guidance de prévention (après incident)
• Maintenir une catégorie "What s New" aussi utilisée comme trigger de fraîcheur pour la récupération IA.

Localisation et accessibilité

Localisation :

• Maintenir un glossaire pour termes produit et labels UI traduits.
• Préférer les instructions texte aux images avec texte intégré.
• Suivre d abord la couverture de traduction des articles à plus fort trafic.

Accessibilité :

• Ajouter texte alt aux images et sous-titres aux vidéos.
• Utiliser titres et listes pour structurer; éviter de porter le sens uniquement par la couleur.
• Garder les étapes scannables et éviter les longs paragraphes.

Alignement support IA

Garder le centre d aide retrieval-friendly :

• Utiliser titres uniques et riches en intention.
• Garder les messages d erreur verbatim et dans des blocs dédiés.
• Ajouter des métadonnées quand la plateforme le permet (zone produit, audience, plan tier, version, last_updated).
• Préférer prérequis explicites et critères de succès explicites.

Définir les règles de sécurité de réponse IA :

• Exiger citations/liens pour réponses factuelles et procédures.
• Poser des questions de clarification quand plan tier, rôle ou version produit affecte les étapes.
• Escalader pour litiges billing, sécurité compte, juridique/conformité et faible confiance.
• Pour les demandes transactionnelles, exiger confirmation explicite avant actions irréversibles.

Maintenir un set d évaluation pour IA et recherche :

• Top 50 recherches et leurs articles de destination attendus
• Top 50 tickets et la réponse self-service minimale viable
• Un set de requêtes de modes d échec (ambiguës, contexte manquant, sensibles aux politiques)

Cadence opérationnelle

Hebdomadaire :

• Revoir les recherches zéro résultat principales et ajouter/retitrer du contenu.
• Revoir les articles "fort trafic + faible utilité" et réécrire un batch.
• Auditer les escalades IA pour identifier gaps de contenu et échecs sécurité.

Mensuelle :

• Rafraîchir captures et labels UI pour les catégories à plus fort trafic.
• Revoir les meilleures opportunités de déflexion depuis les tags ticket.
• Valider la couverture des événements analytics et la santé dashboard.

Trimestrielle :

• Audit taxonomie (prolifération catégories, doublons, navigation cassée).
• Nettoyage du contenu et des redirects.
• Revue de gouvernance (owners, SLAs, playbooks d escalade).

Métriques & optimisation

Suivi KPI, configuration analytics et stratégies d optimisation pour centres d aide.

Contenu

• Framework des métriques cœur
• Calcul ROI
• Setup analytics
• Analytics de recherche
• Performance du contenu
• A/B Testing
• Analyse feedback
• Playbook d optimisation
• Benchmarking
• Alerting & monitoring

Framework des métriques cœur

KPIs primaires

KPIs secondaires

Calcul ROI

Analyse coût-bénéfice

SELF-SERVICE ROI MODEL

Costs:
- Platform subscription: $XXX/month
- Content creation: $XXX/month
- Maintenance: $XXX/month
Total monthly cost: $XXXX

Savings:
- Average cost per ticket: $13
- Tickets deflected: X,XXX/month
- Deflection savings: $XX,XXX/month

Net ROI:
Monthly savings - Monthly cost = Net benefit
(Net benefit / Cost) x 100 = ROI %

EXAMPLE

Platform: $500/month
Content: $1,000/month
Maintenance: $500/month
Total cost: $2,000/month

Deflected tickets: 2,000/month
Cost per ticket: $13
Deflection savings: $26,000/month

Net benefit: $24,000/month
ROI: 1,100%

Coût par résolution

CHANNEL COST COMPARISON

| Channel | Avg. Cost | Resolution Time |
|---------|-----------|-----------------|
| Phone | $15-25 | 8-12 min |
| Email | $10-15 | 24-48 hours |
| Live Chat | $8-12 | 5-10 min |
| AI Chatbot | $0.50-2 | 1-3 min |
| Self-Service | $0.10-0.50 | User-controlled |

TARGET: Maximize self-service, minimize phone

Setup analytics

Configuration Google Analytics 4

// GA4 Event Tracking for Help Center

// Article view
gtag('event', 'article_view', {
  article_id: '12345',
  article_title: 'How to Reset Password',
  category: 'Account',
  content_type: 'how-to'
});

// Search performed
gtag('event', 'search', {
  search_term: 'password reset',
  results_count: 5
});

// Article feedback
gtag('event', 'article_feedback', {
  article_id: '12345',
  feedback_type: 'helpful', // or 'not_helpful'
  feedback_text: 'Optional comment'
});

// Contact support clicked
gtag('event', 'contact_support', {
  source_article: '12345',
  contact_method: 'chat'
});

Événements clés à suivre

ESSENTIAL EVENTS

Page/Article level:
- article_view (with metadata)
- scroll_depth (25%, 50%, 75%, 100%)
- time_on_page
- related_article_click
- external_link_click

Search:
- search_performed
- search_result_click
- zero_results
- search_refinement

Feedback:
- helpful_yes
- helpful_no
- feedback_submitted
- escalation_to_support

AI/Chatbot:
- chatbot_opened
- chatbot_message_sent
- chatbot_resolved
- chatbot_escalated

Template dashboard

HELP CENTER DASHBOARD

Overview Section:
Self-Service Rate: 72%
Deflection: 65%

Search Performance:
Searches today: 1,234
Success rate: 78%
Zero results: 4.2%
Top searches: password, pricing, api

Content Health:
Total articles: 156
Updated <30 days: 45 (29%)
Low-rated (<3/5): 12
High-traffic, low-rated: 5 (priority)

Trend Chart:
[Line chart: tickets, KB views, search success rate]

Analytics de recherche

Métriques de performance recherche

SEARCH METRICS

Volume:
- Total searches/day
- Unique searchers
- Searches per session

Quality:
- Click-through rate (CTR)
- Position of clicked result
- Refinement rate (search again)

Gaps:
- Zero-result queries
- Low-CTR queries
- High-exit searches

ZERO-RESULT ANALYSIS

Weekly review process:
1. Export zero-result queries
2. Group by topic/intent
3. Prioritize by volume
4. Actions:
   - Create new article
   - Add synonyms
   - Update titles
   - Add redirects

Actions d optimisation recherche

Performance du contenu

Modèle de score article

ARTICLE HEALTH SCORE (0-100)

Components:
- Helpfulness rating: 30 points
- Traffic volume: 20 points
- Engagement (time on page): 15 points
- Freshness: 15 points
- Search performance: 10 points
- Link health: 10 points

SCORING EXAMPLE

Article: "How to Reset Password"

Helpfulness: 85% positive -> 25/30 points
Traffic: Top 10% -> 20/20 points
Engagement: 3.5 min avg -> 12/15 points
Freshness: Updated 2 months ago -> 12/15 points
Search: #2 result for "password" -> 8/10 points
Links: All working -> 10/10 points

Total Score: 87/100 (Healthy)

Framework d audit contenu

QUARTERLY AUDIT PROCESS

1. Export all articles with metrics
   - Views (30/90/365 days)
   - Helpfulness rating
   - Last updated date
   - Ticket escalations

2. Categorize by action needed

   OK Healthy (score >70):
   - No action needed
   - Review in 6 months

   Medium Needs attention (50-70):
   - Update content
   - Improve visuals
   - Check accuracy

   Critical Critical (<50):
   - Major rewrite
   - Consider archive
   - Urgent if high-traffic

3. Prioritize by impact
   High traffic + low score = Priority 1
   Low traffic + low score = Consider archive

4. Track improvements
   Before/after metrics per article

Analyse des gaps contenu

IDENTIFYING GAPS

Data sources:
- Zero-result searches
- High-volume support tickets
- User feedback comments
- Sales/success team input
- Product release notes

PROCESS

1. Collect gap signals (weekly)
2. Categorize by topic
3. Score by impact:
   - Ticket volume reduction potential
   - User demand (search volume)
   - Strategic importance

4. Create backlog
5. Prioritize creation

GAP TEMPLATE

Topic: [Gap topic]
Evidence: [Data showing need]
Impact: [High/Medium/Low]
Effort: [Hours to create]
Priority: [P1/P2/P3]
Assigned: [Author]
Due: [Date]

A/B Testing

Quoi tester

TESTABLE ELEMENTS

Titles:
- Question vs. statement
- Verb-first vs. noun-first
- Short vs. descriptive

Content:
- Steps count (5 vs. 10)
- Video vs. text
- Screenshots vs. GIFs

Layout:
- TOC position
- Related articles placement
- CTA button position

Search:
- Result ordering
- Snippet length
- Filter options

Framework A/B test

TEST STRUCTURE

1. Hypothesis
   "Changing [element] from [A] to [B]
   will improve [metric] by [X]%"

2. Success metric
   Primary: [e.g., CTR, helpfulness]
   Secondary: [e.g., time on page]

3. Sample size
   Use calculator for statistical significance
   Minimum: 1,000 views per variant

4. Duration
   Minimum: 2 weeks
   Account for weekly patterns

5. Analysis
   - Statistical significance (p < 0.05)
   - Practical significance (>5% lift)
   - Segment analysis

EXAMPLE TEST

Hypothesis: "How to" prefix increases CTR
Control: "Reset Your Password"
Variant: "How to Reset Your Password"
Metric: Click-through from search
Duration: 2 weeks
Result: +12% CTR (p=0.02) -> Implement

Analyse feedback

Méthodes de collecte feedback

FEEDBACK TYPES

Binary:
"Was this helpful?" [Yes] [No]
- Simple, high response rate
- Limited insight

Rating scale:
"Rate this article" 4/5
- More nuanced
- Moderate response rate

Open text:
"How can we improve this?"
- Rich insight
- Low response rate

Inline feedback:
Highlight -> "Is this unclear?"
- Contextual
- High-quality signal

BEST PRACTICE

Combine:
1. Binary (always show)
2. Follow-up question (on "No")
3. Optional text (for details)

Traitement du feedback

FEEDBACK WORKFLOW

Daily:
- Review new feedback
- Flag urgent issues
- Categorize comments

Weekly:
- Analyze patterns
- Update priority articles
- Report to team

Monthly:
- Trend analysis
- Process improvements
- Content planning input

CATEGORIZATION

- Accuracy issue (content wrong)
- Completeness (missing info)
- Clarity (confusing)
- Outdated (needs update)
- Praise (positive)
- Off-topic (ignore)

Playbook d optimisation

Quick Wins (<1 heure chacun)

IMMEDIATE IMPACT ACTIONS

1. Fix broken links
   - Run link checker
   - Update or remove

2. Add missing screenshots
   - High-traffic how-to articles
   - Error message articles

3. Update dates
   - "Last updated" timestamps
   - Version numbers

4. Add search synonyms
   - Top zero-result queries
   - Common misspellings

5. Improve titles
   - Add action verbs
   - Match search queries

Effort moyen (1 jour chacun)

SIGNIFICANT IMPROVEMENTS

1. Rewrite low-rated articles
   - Address feedback themes
   - Add visual aids
   - Simplify language

2. Create missing content
   - Top 5 zero-result queries
   - Frequent ticket topics

3. Consolidate duplicates
   - Merge similar articles
   - Set up redirects

4. Improve navigation
   - Update category structure
   - Add cross-links
   - Improve breadcrumbs

Projets stratégiques (1 semaine+)

TRANSFORMATIONAL CHANGES

1. AI integration
   - Implement chatbot
   - Set up RAG pipeline
   - Configure escalation

2. Content redesign
   - New templates
   - Consistent formatting
   - Visual refresh

3. Search overhaul
   - Semantic search
   - Personalization
   - Federated search

4. Analytics upgrade
   - Custom dashboards
   - Automated alerts
   - Predictive analytics

Benchmarking

Benchmarks industrie

BENCHMARK RANGES

Self-Service Rate:
- Low: <40%
- Average: 50-65%
- Best-in-class: >75%

Ticket Deflection:
- Low: <20%
- Average: 30-45%
- Best-in-class: >55%

Search Success:
- Low: <60%
- Average: 70-80%
- Best-in-class: >85%

CSAT (KB):
- Low: <70%
- Average: 75-82%
- Best-in-class: >88%

NOTE: Benchmarks vary by industry
- B2B SaaS: Higher self-service expected
- E-commerce: Lower (simpler queries)
- Enterprise: Variable by product complexity

Analyse concurrentielle

COMPETITIVE INTEL CHECKLIST

Analyze competitor help centers:

Structure:
- Category organization
- Article types
- Navigation patterns
- Search prominence

Content:
- Writing style
- Visual approach
- Depth of content
- Update frequency

Features:
- AI chatbot presence
- Community forums
- Video content
- Interactive guides

UX:
- Mobile experience
- Load time
- Accessibility
- Personalization

Document findings:
- What they do better
- What we do better
- Opportunities to differentiate

Alerting & monitoring

Configuration alertes

AUTOMATED ALERTS

Critical (immediate):
- Zero-result rate >10%
- Helpfulness <60%
- Site down/errors

Warning (daily digest):
- Traffic drop >20% WoW
- New low-rated articles
- Stale content (>6 months)

Info (weekly summary):
- Top performing content
- Trending searches
- Feedback themes

ALERT TEMPLATE

Subject: [Severity] Help Center Alert: [Issue]

What: [Description of issue]
Impact: [Metric change]
Affected: [Articles/pages]
Action: [Recommended fix]
Link: [Dashboard/article link]

Automatisation health check

WEEKLY AUTOMATED CHECKS

- Broken link scan
- Image loading verification
- Search functionality test
- Chatbot response test
- Mobile rendering check
- Load time measurement
- SSL certificate validity
- Analytics tracking verification

MONTHLY AUTOMATED REPORTS

- Content freshness report
- Search performance summary
- Feedback trend analysis
- Traffic comparison (MoM, YoY)
- Top/bottom performers
- Gap analysis update

name: product-help-center description: Concevoir ou auditer des centres d aide et bases de connaissances AI-first. À utiliser pour taxonomie, templates d articles, setup RAG ou planification de chatbot support.

Design de centre d aide

Concevoir des centres d aide, bases de connaissances, FAQ et supports d apprentissage AI-first.

Cette skill reflète le passage des portails d aide statiques vers des systèmes self-service alimentés par IA, embarqués et personnalisés.

Workflow (ordre par défaut)

1. Définir périmètre et contraintes
   • Audience/personas, zones produit, versioning produit, canaux (web/in-app), exigences conformité, besoins de localisation.
2. Inventorier la connaissance actuelle
   • Principaux tickets, recherches, articles, raisons d escalade et owners connus du contenu.
3. Construire l architecture d information
   • Structure de catégories, tagging, navigation, stratégie URL et liens internes.
4. Standardiser le contenu
   • Types d articles, templates, règles d écriture compatibles IA et standards visuels.
5. Instrumenter et mesurer
   • KPIs, event tracking, dashboards et logging des requêtes de recherche.
6. Ajouter le support IA en sécurité
   • Réponses retrieval-first, citations, seuils de confiance, règles d escalade et garde-fous transactionnels.
7. Opérer la connaissance
   • Gouvernance, détection de fraîcheur, updates liés aux releases et optimisation continue.

Sorties attendues (adapter à la demande) :

• Carte de taxonomie du centre d aide + schéma de tags
• Backlog top 20 articles (par impact) + templates
• Spec analytics (événements + KPIs dashboard)
• Spec support IA (sources RAG, seuils d escalade, règles sécurité)
• Cadence opérationnelle (owners + planning de revue)

Référence rapide

Matrice de décision des types de contenu

Choix de plateforme (vérifier prix et limites de plan)

Voir references/platform-guides.md pour les notes setup/migration et data/sources.json pour les sources de comparaison curées.

Bonnes pratiques 2025-2026

Changements clés

Évitez de citer des statistiques dures sans vérification; rafraîchissez tendances et benchmarks via data/sources.json si nécessaire.

Principes AI-first

1. Résolution agentique — l IA exécute des tâches (remboursements, réservations, updates), pas seulement répond
2. Compréhension sémantique — recherche par intention, pas matching mots-clés
3. Assistance proactive — faire remonter l aide avant que les utilisateurs demandent
4. Fraîcheur du contenu — détecter automatiquement le contenu obsolète, suggérer des updates
5. Synthèse multi-source — tirer depuis docs, tickets, Slack, release notes
6. IA riche en mémoire — conserver le contexte entre sessions pour un support personnalisé

Tendances émergentes (2026)

Architecture du centre d aide

Règles de structure de catégories

HIERARCHY LIMITS
- Maximum depth: 2-3 levels
- Top-level categories: 5-9 (cognitive load principle)
- Articles per category: 10-20 (scannable)
- Avoid: Deep nesting, internal org structure

Catégories top-level recommandées

STANDARD CATEGORIES (adapt to product)
1. Getting Started        — First-run, setup, quick wins
2. [Core Feature 1]       — Primary use case
3. [Core Feature 2]       — Secondary use case
4. Account & Billing      — Settings, payments, security
5. Integrations           — Third-party connections
6. Troubleshooting        — Common issues, error codes
7. API & Developers       — Technical documentation
8. What's New             — Changelog, releases

Patterns de navigation

• Breadcrumbs — toujours montrer la position dans la hiérarchie
• Articles liés — 3-5 liens contextuellement pertinents
• Prochaines étapes — guider vers l action logique suivante
• Recherche proéminente — au-dessus de la ligne de flottaison, toujours visible
• Articles populaires — faire remonter le contenu à fort trafic

Types d articles (garder l ensemble réduit)

• How-To : réalisation de tâche, 3-10 étapes
• Troubleshooting : symptômes -> causes -> solutions
• FAQ : réponses rapides avec liens vers docs plus profondes
• Conceptuel : expliquer termes et modèles mentaux
• Référence : specs précises (tableaux, limites, codes d erreur)

Utilisez les templates copier-coller dans references/article-templates.md.

Patterns d intégration IA

Architecture chatbot

MODERN AI SUPPORT FLOW (2025)

User query
  -> Intent detection (semantic understanding)
  -> RAG retrieval (KB + tickets + docs)
  -> Response and action (answer and/or execute task)
  -> Escalation check (confidence below threshold?)
  -> Human agent (if needed)

Capacités IA agentiques (2025-2026)

Contenu pour consommation IA

AI-FRIENDLY WRITING RULES

DO:
- Clear headings with keywords
- Structured data (tables, lists)
- Explicit step numbering
- Error messages verbatim
- Unique article titles

DON'T:
- Ambiguous pronouns
- Implicit assumptions
- Marketing fluff in support content
- Duplicate content across articles

Voir references/ai-integration.md pour setup RAG, évaluation et patterns d escalade.

Métriques & KPIs

Métriques cœur

Métriques de santé du contenu

FRESHNESS INDICATORS
- Last updated > 6 months -> Review required
- Last updated > 12 months -> Likely stale
- No views in 90 days -> Consider archive
- High bounce rate -> Content mismatch

QUALITY INDICATORS
- Thumbs down > 20% -> Rewrite needed
- Escalation after viewing -> Content gap
- Search -> immediate exit -> Title mismatch

Calcul ROI

SELF-SERVICE ROI FORMULA

Monthly Savings = (Deflected Tickets x $13) - Platform Cost

Example:
- 1,000 deflected tickets/month
- $13 average agent cost
- $500 platform cost
- ROI = ($13,000 - $500) = $12,500/month

Voir references/metrics-optimization.md pour instrumentation, dashboards et playbooks d optimisation.

Apprentissage & onboarding

Patterns d aide in-app

Bonnes pratiques tutoriels (2025)

VIDEO TUTORIALS
- Length: 2-4 minutes (40% higher completion)
- Format: Screen recording + voiceover
- Chapters: Clickable sections
- Captions: Always include (accessibility)

INTERACTIVE GUIDES
- Click-through walkthroughs
- Sandbox environments
- Progress saving
- Skip option for experienced users

Voir references/learning-paths.md pour design de séquences onboarding, accessibilité et mesure.

Knowledge Operations (2026)

Opérez le centre d aide comme un produit :

• Assigner des owners par catégorie et par article top; définir cadence de revue et SLAs d update.
• Utiliser release notes, rapports d incident et tendances tickets comme triggers automatiques de mise à jour contenu.
• Utiliser signaux de fraîcheur (sorties de recherche, escalade après vue article, downvotes) pour prioriser les réécritures.

Voir references/knowledge-ops.md pour gouvernance, workflows et checklists.

Checklist d implémentation

Phase 1 : Fondation (semaine 1-2)

REQUIRED:

• Choisir la plateforme (Zendesk/Intercom/Freshdesk)
• Définir la structure de catégories (5-9 top-level)
• Créer les templates d articles pour chaque type
• Configurer le tracking analytics
• Configurer les paramètres de recherche

Phase 2 : Contenu (semaine 3-4)

REQUIRED:

• Auditer la documentation existante
• Migrer/réécrire les 20 principaux articles
• Ajouter du contenu visuel (captures, GIFs)
• Implémenter les liens internes
• Configurer les redirects depuis les anciennes URLs

Phase 3 : Intégration IA (semaine 5-6)

REQUIRED:

• Activer le chatbot IA
• Configurer RAG/recherche sémantique
• Fixer les seuils d escalade
• Tester les requêtes courantes
• Suivre les taux de résolution

Phase 4 : Optimisation (continu)

REQUIRED:

• Revoir chaque semaine les recherches zéro résultat
• Mettre à jour mensuellement le contenu obsolète
• A/B tester les titres d articles
• Analyser les patterns d escalade
• Étendre à partir des tendances de tickets

Ressources

Protocole de veille tendances

REQUIRED: Quand les utilisateurs posent des questions de recommandation sur centres d aide, bases de connaissances ou plateformes support, lancez une recherche web rapide pour confirmer les tendances actuelles avant de répondre. Préférez les sources marquées add_as_web_search: true dans data/sources.json, plus les docs officielles de toute plateforme recommandée.

Conditions de déclenchement

• "What's the best help center platform?"
• "What should I use for [knowledge base/FAQ/support]?"
• "What's the latest in customer self-service?"
• "Current best practices for [AI support/chatbots]?"
• "Is [Zendesk/Intercom/Freshdesk] still relevant in 2026?"
• "[Zendesk] vs [Intercom] vs [other]?"
• "Best AI chatbot for customer support?"

Recherches requises

1. Search: "help center best practices 2026"
2. Search: "[specific platform] vs alternatives 2026"
3. Search: "AI customer support trends January 2026"
4. Search: "knowledge base platforms 2026"

Quoi rapporter

Après recherche, fournissez :

• Paysage actuel : quelles plateformes/outils support sont populaires MAINTENANT
• Tendances émergentes : nouvelles capacités IA, patterns ou plateformes qui gagnent du terrain
• Déprécié/en déclin : approches ou outils perdant en pertinence
• Recommandation : fondée sur des données fraîches, pas seulement sur connaissance statique

Si la recherche web est indisponible, indiquez cette contrainte et poursuivez avec la meilleure guidance statique.

Exemples de sujets (à vérifier par recherche fraîche)

• Plateformes de centre d aide (Zendesk, Intercom, Freshdesk)
• Agents support IA (Fin AI, Ada, Forethought)
• Outils knowledge base (Document360, GitBook, Notion)
• Guidage in-app (UserPilot, Pendo, Chameleon)
• Capacités IA self-service et taux de résolution

Vérification des faits

• Utiliser recherche web/web fetch pour vérifier les faits externes actuels, versions, prix, échéances, réglementations ou comportements de plateforme avant les réponses finales.
• Préférer les sources primaires; rapporter liens sources et dates pour les informations volatiles.
• Si l accès web est indisponible, indiquer la limite et marquer la guidance comme non vérifiée.

Patterns de taxonomie

Patterns d architecture d information pour centres d aide et bases de connaissances.

Contenu

• Règles de hiérarchie de catégories
• Structures de catégories standard
• Organisation centrée utilisateur
• Stratégies de tagging
• Optimisation de recherche
• Patterns de navigation
• Stratégie de liens croisés
• Déduplication de contenu
• Structure URL

Règles de hiérarchie de catégories

Limites de profondeur

HIERARCHY BEST PRACTICES

Maximum depth: 3 levels
Optimal depth: 2 levels
Top-level categories: 5-9 (Miller's Law)
Articles per category: 10-20

BAD: Products > Software > Desktop > Windows > Settings > Display
GOOD: Settings > Display Settings

Principe de charge cognitive

Les utilisateurs peuvent garder 7 +/- 2 éléments en mémoire de travail. Appliquez-le à :

Structures de catégories standard

Produit SaaS (B2B)

RECOMMENDED STRUCTURE

1. Getting Started
   |-- Quick Start Guide
   |-- Account Setup
   \-- First Project

2. [Core Feature 1]
   |-- Overview
   |-- How-To Guides
   \-- Best Practices

3. [Core Feature 2]
   |-- Overview
   |-- How-To Guides
   \-- Best Practices

4. Integrations
   |-- Native Integrations
   |-- API
   \-- Zapier/Make

5. Account & Billing
   |-- Account Settings
   |-- Team Management
   |-- Billing & Invoices
   \-- Security

6. Troubleshooting
   |-- Common Issues
   |-- Error Messages
   \-- Performance

7. What's New
   |-- Release Notes
   \-- Roadmap

Plateforme e-commerce

RECOMMENDED STRUCTURE

1. Getting Started
   |-- Account Creation
   |-- First Order
   \-- App Download

2. Orders & Shipping
   |-- Track Order
   |-- Shipping Options
   |-- Returns & Exchanges
   \-- Order Issues

3. Payments
   |-- Payment Methods
   |-- Refunds
   |-- Gift Cards
   \-- Payment Issues

4. Account
   |-- Profile Settings
   |-- Addresses
   |-- Password & Security
   \-- Notifications

5. Products
   |-- Size Guides
   |-- Care Instructions
   \-- Availability

6. Loyalty Program
   |-- How It Works
   |-- Points & Rewards
   \-- Member Benefits

Plateforme développeurs

RECOMMENDED STRUCTURE

1. Getting Started
   |-- Quick Start
   |-- Installation
   |-- Authentication
   \-- First API Call

2. Guides
   |-- Core Concepts
   |-- Tutorials
   \-- Best Practices

3. API Reference
   |-- Endpoints
   |-- Authentication
   |-- Rate Limits
   \-- Errors

4. SDKs & Libraries
   |-- JavaScript
   |-- Python
   |-- Ruby
   \-- Go

5. Integrations
   |-- Webhooks
   |-- OAuth
   \-- Third-Party

6. Resources
   |-- Changelog
   |-- Status Page
   \-- Community

Organisation centrée utilisateur

Organiser par objectif utilisateur, pas par fonctionnalité

WRONG (feature-centric)
|-- Dashboard
|-- Reports Module
|-- Settings Panel
|-- API Section

RIGHT (goal-centric)
|-- Track Performance
|-- Analyze Results
|-- Configure Your Account
|-- Build Integrations

Catégories par audience

MULTI-AUDIENCE STRUCTURE

For Users
|-- Getting Started
|-- Daily Tasks
\-- Troubleshooting

For Admins
|-- Setup & Configuration
|-- User Management
|-- Security & Compliance

For Developers
|-- API Reference
|-- SDKs
\-- Webhooks

Catégories par parcours

USER JOURNEY STRUCTURE

Evaluate
|-- Product Overview
|-- Pricing
|-- Comparison Guides

Onboard
|-- Quick Start
|-- Initial Setup
|-- First Success

Use Daily
|-- Core Workflows
|-- Tips & Tricks
|-- Shortcuts

Expand
|-- Advanced Features
|-- Integrations
|-- Team Collaboration

Troubleshoot
|-- Common Issues
|-- Error Reference
|-- Contact Support

Stratégies de tagging

Tags plats (recommandé pour <500 articles)

TAG TYPES

Topic tags: billing, security, api, mobile
Audience tags: admin, user, developer
Content type: how-to, troubleshooting, reference, faq
Product area: dashboard, reports, settings
Difficulty: beginner, intermediate, advanced

Tags hiérarchiques (pour >500 articles)

TAG HIERARCHY

integration/
|-- integration/native
|-- integration/api
|-- integration/zapier
\-- integration/webhooks

billing/
|-- billing/payments
|-- billing/invoices
|-- billing/refunds
\-- billing/subscriptions

Gouvernance des tags

Optimisation de recherche

Synonymes & redirects

SYNONYM MAPPING

User searches -> Canonical term
"password reset" -> "reset password"
"cost" -> "pricing"
"sign up" -> "create account"
"login" -> "sign in"
"delete" -> "remove"
"cancel" -> "unsubscribe"

REDIRECT RULES

/help/billing -> /help/account/billing
/faq -> /help
/support -> /help

Ranking des résultats de recherche

RANKING FACTORS (priority order)

1. Title match (exact)
2. Title match (partial)
3. Heading match
4. Body content match
5. Tag match
6. Popularity (views)
7. Freshness (updated date)

BOOST FACTORS

+50% Getting Started articles (for new users)
+30% Recently updated content
+20% High-rated content
-50% Archived content

Gestion des recherches zéro résultat

ZERO-RESULT STRATEGY

1. Track all zero-result queries
2. Weekly review of top 20 queries
3. Actions:
   - Create new article
   - Add synonyms
   - Update existing article title
   - Add to FAQ

FALLBACK UI

"No results for '[query]'"
- Did you mean: [suggestions]
- Popular articles: [top 3]
- Browse categories: [list]
- Contact support: [link]

Patterns de navigation

Breadcrumbs

BREADCRUMB RULES

Format: Home > Category > Subcategory > Article
Separator: > or /
Clickable: All except current page
Mobile: Collapse to "... > Parent > Current"

EXAMPLE
Help Center > Account > Security > Enable Two-Factor Auth

Articles liés

RELATED ARTICLES LOGIC

Display: 3-5 articles
Position: End of article, sidebar
Selection criteria:
1. Same category (weight: 40%)
2. Shared tags (weight: 30%)
3. User behavior (also viewed) (weight: 20%)
4. Manual curation (weight: 10%)

EXCLUDE
- Current article
- Archived articles
- Different audience level

Prochaines étapes / Call-to-Action

NEXT STEPS PATTERN

After how-to:
-> Related advanced guide
-> Troubleshooting for this feature
-> Video tutorial

After troubleshooting:
-> Contact support (if unresolved)
-> Related how-to
-> Community forum

After conceptual:
-> How-to using this concept
-> API reference
-> Example project

Table des matières

TOC RULES

Show when: Article > 500 words OR > 3 headings
Position: Top of article, sticky sidebar
Depth: H2 and H3 only
Clickable: Smooth scroll to section
Highlight: Current section in view

Stratégie de liens croisés

Règles de liens internes

Maintenance des liens

LINK HEALTH CHECKS

Weekly:
- [ ] Check for broken links (404s)
- [ ] Update redirects for moved content

Monthly:
- [ ] Review orphan pages (no incoming links)
- [ ] Check for circular references
- [ ] Update outdated cross-references

Quarterly:
- [ ] Full link audit
- [ ] Update deprecated content links
- [ ] Review external links

Déduplication de contenu

Éviter les doublons

SINGLE SOURCE OF TRUTH

Problem: Same info in multiple places
Solution: One canonical article + links

EXAMPLE

BAD:
- Article A: "How to reset password" (full steps)
- Article B: "Account security" (same steps inline)
- FAQ: "How do I reset password?" (same steps)

GOOD:
- Article A: "How to reset password" (full steps)
- Article B: "Account security" (link to A)
- FAQ: "How do I reset password?" (link to A)

Patterns de réutilisation de contenu

REUSABLE COMPONENTS

Warnings/Notes:
<!-- include: security-warning.md -->

Common steps:
<!-- include: navigate-to-settings.md -->

Product limits:
<!-- include: plan-limits-table.md -->

IMPLEMENTATION
- Zendesk: Content blocks
- Intercom: Reusable content
- GitBook: Reusable content / includes
- Notion: Synced blocks

Structure URL

Bonnes pratiques URL

URL PATTERNS

Good:
/help/billing/upgrade-plan
/docs/api/authentication
/guides/getting-started

Bad:
/help/article/12345
/kb/cat-billing/sub-payments/art-upgrade
/help/billing_and_payments/how_to_upgrade_your_plan

RULES
- Lowercase only
- Hyphens (not underscores)
- No IDs in URL
- Max 3 levels deep
- Descriptive slugs

Redirects URL

REDIRECT TYPES

301 (Permanent): Content moved forever
302 (Temporary): Testing, A/B
Canonical: Duplicate content prevention

WHEN TO REDIRECT
- Article renamed
- Category restructured
- Content merged
- Old URLs bookmarked/linked externally