Service · Audit 01.

I measure what AI agents see of you.
Then you decide what we do with it.

When a client asks ChatGPT, Claude, Perplexity or Gemini to recommend a house like yours, what comes out? Your words? Your numbers? Your narrative? Or a competitor's? The audit runs hundreds of structured queries, crosses the four agents on six independent dimensions, and delivers five argued hypotheses and a prioritised action plan.

The deliverable, in four extracts

Sixty pages. Six dimensions. Five argued hypotheses. A prioritised action plan.

Glimpses from two real audits — an independent Parisian fashion house et a premium ready-to-wear house — anonymisés. Données et constats inchangés.

Section 01 Audit · Parisian fashion house

Executive summary

Présence globale 92% House mentioned in responses
Spontaneous Discovery 52% Queries without brand name
Inaccuracies detected 8 Priority corrections
Co-cited competitors 8 Brands captured by LLMs

Radical asymmetry between dimensions where the house is named by the prospect (above 80) and those where it must be found (33 to 53). The agentic acquisition pipeline is the territory to reclaim.

Section 03 Agentic cartography · 6D

Composite score + six independent sub-scores

44 /100 Grade C
  • Discovery
    34
  • Comparison
    53
  • Knowledge
    42
  • Purchase
    45
  • Ethics
    57
  • Loyalty
    46

The shape of the score tells more than its value. A house at 44/100 with Discovery at 34 and Ethics at 57 is not the same house as another flat at 44/100.

Section 05 Signaux détectés

Three findings structure the action territories

  1. 01

    Failing agentic recruitment.

    Quand le prospect cherche par signaux distinctifs (sans nommer la maison), Perplexity ne la retrouve que 33% du temps. ChatGPT 60%, Claude 66%, Gemini 80%. L'angle mort est précisément là où la conversion non-existante se joue.

  2. 02

    Convergent factual hallucinations.

    The 4 LLMs hallucinent collectivement la date de fondation de la maison. Trois agents proposent une date erronée commune, le quatrième en propose une autre — la date correcte n'apparaît dans aucune réponse. Signal net d'un référentiel public mal indexé sur un fait basique.

  3. 03

    Operational lead is invisible.

    Replatforming e-commerce livré, migration ERP cloud aboutie, app marché Asie avec carte de fidélité : chantiers documentés publiquement par les partenaires. None of the 4 LLMs returns them spontaneously. Asymétrie entre l'avance opérationnelle réelle et la densité documentaire restituable.

Section 09 Plan d'action priorisé

Quick wins, structural projects, narrative doctrine

  1. Étape 1

    Editorial quick wins

    • Page « Histoire de la maison » enrichie de dates pivots, JSON-LD Organization + foundingDate
    • Structured FAQ (positioning, signature, independence, governance)
    • llms.txt publié, priorités décrites explicitement
  2. Étape 2

    Structural projects

    • Refonte des pages produit phares (Schema Product + Brand + reviews)
    • Communiqué éditorial signé direction sur l'avance opérationnelle (replatforming, ERP, Asie)
    • Sourcing externe : articles tiers thématiques, plan d'amorçage Wikipédia
  3. Étape 3

    Agentic narrative doctrine

    • Agentic brand platform: narrative invariants, what agents are allowed to say
    • Re-test sur les requêtes initiales, mesure du delta sur les six sub-scores
    • Recurring editorial brief sent to priority third-party media

The deux audits référencés ont été livrés en avril et May 2026. Noms de marques, concurrents nommément cités, dates précises et noms de partenaires technologiques sont anonymisés ou agrégés. Données et constats restent ceux des audits réels.

The six audited dimensions

Six independent lenses. Six precise business questions.

Chaque dimension correspond à un type de requête réel d'un prospect ou d'un client. Read la performance dimension par dimension permet de voir où la maison brille et où elle décroche — un score moyen masque cette asymétrie.

DÉCOUVERTE COMPARAISON CONNAISSANCE ACHAT ÉTHIQUE FIDÉLISATION recrutement non-intentionnel positionnement vs pairs histoire, codes, doctrine modalités d'acquisition engagements, perception cote, programme, SAV
Six axes, six questions clientes, six lectures. Le radar est l'épure — le rapport en livre la silhouette pour votre maison.
  1. 01

    Discovery

    « Are you cited when searching your category? »

    25%
  2. 02

    Comparison

    « What does the agent say when comparing you to a competitor? »

    15%
  3. 03

    Knowledge

    « What the agent knows — and what it invents. »

    20%
  4. 04

    Purchase

    « Does the agent put you on the transactional path? »

    15%
  5. 05

    Ethics

    « Do your CSR commitments arrive intact? »

    15%
  6. 06

    Loyalty

    « What does the agent recommend once the first sale is closed? »

    10%
Best of both worlds

The power of the algorithm. The knowledge of the sector. Combined.

Un audit GEO automatisé livre un dashboard. Un consultant senior livre une opinion. Sur le luxe, ni l'un ni l'autre ne suffit. NarrativeGuard 2.0 moissonne ce qu'aucun œil humain ne pourrait croiser à cette échelle ; l'expertise sectorielle lit ce qu'aucun algorithme ne saurait interpréter. La combinaison est le produit.

La puissance de l'algorithme

NarrativeGuard 2.0

The proprietary LUXE ÆTERNAI agent. It runs the queries, crosses the four agents across multiple runs, scores every answer on a sales-ready weighted grid (recommendation, position, accuracy, actionability, sentiment, differentiation). What no human hand could reproduce at this scale.

  • Pose les requêtes structurées sur quatre LLMs, runs multiples
  • Score chaque réponse selon six critères pondérés propriétaires
  • Crosses the agents, detects convergent and divergent patterns
  • Archives the raw, timestamped dataset, reusable for your subsequent audits
La connaissance du secteur

Twenty years of digital luxury

Twenty years of digital in luxury (LVMH, Kering, Razorfish, WIDE, Australie GAD). Seven years in Gen AI. I decide what the numbers mean for your house — not another. Five argued hypotheses, a prioritised plan: what no algorithm can formulate alone.

  • Reads the gaps between what you build and what the agents retain
  • Formulates five argued, sourced, ranked hypotheses
  • Sizes each workstream in effort × impact, proposes an owner type
  • Delivers to the Exec Board, defends the arbitrations, supports the follow-up

« The algorithm sees, expertise decides. Without the first, we guess. Without the second, we benchmark. Both together: that's the audit. »

The flow

Three steps. Your inputs feed the audit, not the other way.

  1. 01 Scoping

    Your inputs define the scope.

    Workshop with your team (CMO, CX Director, e-commerce Director). You validate the personas addressed, the questions asked to the agents, the competitors benchmarked. Without this step, the audit measures the wrong ground.

    • FormatVideo workshop
    • DeliverableSigned scoping note
  2. 02 Collection

    NarrativeGuard 2.0 takes over the harvest.

    Four agents, multiple runs, proprietary scoring. Raw responses archived, aggregated scoring, full documented dataset. Mechanical phase, no client interaction.

    • Scope4 LLMs, multiple runs
    • DeliverableFull raw dataset
  3. 03 Analysis & delivery

    Expertise turns data into decisions.

    Critical reading of the dataset, formulation of the five argued hypotheses, prioritised action plan. Delivery to the executive committee: Q&A format, defence of the arbitrations, follow-up included.

    • DeliveryExec Board presentation + follow-up
    • Deliverable60-page document + action plan
The differential

Pourquoi six dimensions, et pas cinq comme on lit ailleurs.

La méthodologie publique des audits GEO se cale sur 5 dimensions héritées du SEO. Sur le luxe, deux terrains méritent d'être isolés : l'Ethics (le récit RSE peut dérailler indépendamment du reste) et la Loyalty (le programme client est rarement bien représenté côté agent). The noyer dans une moyenne, c'est rater deux signaux que les directions générales lisent en premier.

Audit GEO 5DNarrativeGuard 2.0
DiscoveryInclusInclus
ComparisonInclusInclus
KnowledgeInclusInclus
PurchaseInclusInclus
Ethics & RSEDiluted in KnowledgeFull dimension in its own right
LoyaltyHors scopeFull dimension in its own right
PondérationArithmetic meanPondération vente-ready
Robustesse1 runRuns multiples sur fenêtres distinctes
Lecture critiqueSortie automatiséeExpertise humaine + agent
Investment

Un ticket d'entrée clair. Pas de retainer. Pas de markup sur les LLMs.

Format standard

Audit complet, scope unique.

On quote

Inclus

  • 1 marque, 6 dimensions, 4 LLMs
  • 3 concurrents directs benchmarkés
  • 60-page document delivered
  • Executive committee delivery
  • Call de calibration post-restitution
Discuss ce format →
Bespoke

Périmètres élargis, devis à façon.

On quote

Élargissements possibles

  • Multi-marques (groupe, portefeuille)
  • Multi-geography (FR + EU + US + APAC)
  • Dimensions additionnelles (B2B, marketplace, seconde main)
  • Re-test post-mise en œuvre (mesure d'impact)
  • Atelier doctrine post-audit
  • Présentation board étendue
Cadrer un sur-mesure →
After the audit

Trois suites possibles. Aucune imposée.

→ Suite 1

You internalise.

The calibration call is included. You keep the dataset, the report, the plan. Your teams execute. Many houses run the audit without immediate follow-up, and come back later to measure impact.

→ Suite 2

You entrust me with the doctrine.

Offer AI Doctrine & Roadmap. We translate the audit's hypotheses into an agentic brand platform — what narrative agents are allowed to carry, what client experience we delegate to them, what data we open.

→ Suite 3

You take an advisory.

Offer Monthly Retainer. Regular strategic presence to steer plan execution, arbitrate workstreams, recalibrate quarterly. One single point of contact, committed over time.

Before you go

A framed conversation. Not a discovery call.

You describe your house, your scope, your challenges. I tell you if the audit is the right entry point — or if something else should come first. Personal answer.

Discuss this audit → Free diagnostic →
Frequently asked questions

Six questions, six straight answers.

The NarrativeGuard 2.0 methodology isolates Ethics and Loyalty as standalone dimensions. In luxury, the CSR narrative and the client programme are two terrains where the AI agent can derail brand perception independently from the rest. Models with 4 or 5 dimensions drown them in averages — we surface them.
ChatGPT-5 (OpenAI), Claude Sonnet 4.6 (Anthropic), Perplexity Pro and Gemini 2.5 (Google). These four cover >95% of agentic traffic observed on luxury queries in France and Europe (Similarweb agent traffic 2026, Datos.org). On request: Mistral Le Chat or DeepSeek for specific scopes.
Queries are submitted across multiple distinct time windows to neutralise the LLM-side A/B testing effect. Raw responses archived, scoring aggregated. If an agent responds unstably beyond a threshold, it's a qualified signal in the report — not hidden noise.
The diagnostic takes up 25% of the document. The remaining 75%: five argued hypotheses and a prioritised action plan. Each workstream is sized (effort × impact), assigned to an owner type, measurable. An audit that doesn't translate into 3 Exec Board decisions is worthless — that's the angle.
The raw dataset and report belong to you. No reuse, no anonymisation for public benchmark without agreement. NDA signed before scoping. The queries submitted to LLMs are generic client questions (never prompt injection or attempts to extract protected data).
Even better — you have the baseline. NarrativeGuard 2.0 + the expertise cover the missing dimensions (Ethics, Loyalty), introduce a proprietary critical reading and multi-run robustness. The delta vs your prior audit often delivers the most actionable insight in the report. Same price.