QueryArc
    AuditOptimizationMonitoringPricing
    Check my AI visibility

    AI visibility audit methodology

    How QueryArc measures brand visibility across 5 AI assistants (ChatGPT, Claude, Gemini, Perplexity, Grok), reduces variance with repeat runs, and keeps results traceable back to raw outputs.

    Rule-based scoringMulti-engine testing3 runs per promptStructured human diagnosisStored run recipeRaw output traceability
    Controlled question setMulti-engine testingThree runs per promptRule-based scoringMetric calculationsReporting layersTraceabilityHuman diagnosis frameworkStored run recipeVariance"Together" explanationLimitationsSummary

    1.Controlled question set

    Each audit uses a structured prompt pack designed around real buyer-intent queries. Prompts are clearly defined, versioned, stored, and reused for reruns. This prevents question drift and allows comparable measurement over time.

    prompt_pack: v1

    total_prompts: 10

    non_brand: 8

    brand_name: 2

    The full prompt list is included in the report.

    2.Multi-engine testing

    Audits are run across multiple AI engines. Results are shown separately for each engine. We do not blend or average engine results without disclosure.

    ChatGPT

    Runs per prompt: 3

    Reported separately

    Claude

    Runs per prompt: 3

    Reported separately

    Gemini

    Runs per prompt: 3

    Reported separately

    Perplexity

    Runs per prompt: 3

    Reported separately

    Grok

    Runs per prompt: 3

    Reported separately

    Engine-level reporting is never overridden by aggregate views.

    3.Three runs per prompt

    Each prompt is executed three times per engine. AI models are probabilistic systems and may produce slightly different responses on repeated runs. Running multiple times reveals stable patterns.

    Prompt A
    Answer consistency: 100%

    Only successfully completed runs are included in denominators.

    4.Rule-based scoring system

    Every AI response is evaluated using fixed detection rules. No human manually adjusts scores.

    Scoring rules — audit_v1.0

    5.Metric calculations

    Percentages and rates are calculated using a single formula. No estimates, sampling adjustments, or manual overrides are applied.

    Matching runs ÷ Total valid runs

    Only successfully completed runs are included.

    6.Model-level and question-level reporting

    Reports include multiple layers of analysis. This shows exactly where outcomes differ across prompts and models.

    Executive snapshot
    ↓
    Model-level breakdown
    ↓
    Question-level breakdown
    ↓
    Representative excerpts
    ↓
    Raw outputs
    ↓
    Scored dataset

    7.Representative excerpts and traceability

    Reports include representative excerpts from AI responses. Full raw outputs and scored datasets are preserved in technical appendices. This allows complete traceability from summary metrics back to original model responses.

    Nothing is manually interpreted before scoring.

    8.Human diagnosis framework

    After quantitative scoring is complete, results are interpreted using a structured diagnosis framework. Diagnosis does not modify mention rates, endorsement rates, ranking metrics, or comparison outcomes. All numerical values remain rule-based.

    Scoring layer

    Rule-based, deterministic

    Diagnosis layer

    Structured interpretation

    Diagnosis does not modify numerical results.

    Diagnosis categories

    Blueprint mapping

    Low non-brand mention → Category association gap → Category landing page

    Diagnosis explains patterns. It does not alter scoring outputs.

    9.Stored run recipe

    Each audit stores its run configuration. When reruns are performed, the same configuration is used to maintain comparability.

    prompt_pack: v1

    models: [ChatGPT, Claude, Gemini, Grok, Perplexity]

    timestamp: 2025-01-15T09:30:00Z

    settings: stored

    Reruns use the same stored configuration.

    10.Understanding variance

    AI model behavior can vary between runs and can evolve over time due to provider updates. No single response determines a conclusion.

    AI models are probabilistic systems.

    • 3-run execution reduces volatility
    • Consistency is measured
    • Behavior reflects time of testing

    11.What "together" means

    In some views, results may be shown "together." This means all valid runs across engines are combined into a single aggregated dataset.

    GPT
    Claude
    Gemini
    Pplx
    Grok
    →
    Aggregated dataset

    Engine-level reporting remains available.

    12.Limitations

    AI systems evolve over time. Outcomes may change as models are updated.

    • We do not guarantee rankings
    • We do not guarantee placement in AI responses
    • We do not claim privileged access to AI systems
    • We do not modify or manipulate model behavior
    • We do not manually adjust scoring

    13.Summary

    QueryArc uses:

    • A controlled question set
    • Multi-engine testing
    • Three runs per prompt
    • Fixed rule-based scoring
    • Structured human diagnosis
    • Model-level reporting
    • Question-level reporting
    • Raw output traceability
    • Stored run configurations for reruns

    This produces auditable, reproducible, and interpretable AI visibility metrics.

    Ready to see how AI talks about you?

    Check my AI visibility

    Free — no credit card · No login required · Results in under 60 seconds

    The loop

    • Audit
    • Optimization
    • Monitoring
    • Pricing

    How it works

    • The full cycle
    • Methodology
    • Sprint
    • Who It's For
    • Security

    Company

    • About
    • Contact
    • Privacy
    • Terms

    Learn

    • AI Visibility
    • AI Visibility Audit
    • ChatGPT Answers
    • AI SEO
    • LLM SEO
    • Arc Rank

    Check your AI visibility in under a minute

    Check my AI visibility

    Free — no credit card · No login required

    Check my AI visibility

    Free — no credit card · Results in under 60 seconds

    QueryArc

    AI Visibility Audits for ChatGPT, Claude, Gemini & AI Search

    Private reports · No ranking guarantees · Stored run recipe included