Methodology
Transparent by design. Auditable by default.
Enterprise procurement teams asked us how we measure. So we wrote it down.
Prompt construction
We generate commercial-intent prompts using LLM-assisted templating, then human-curate for category fit. Each brand gets 200–2,000 prompts spanning awareness, comparison, and purchase intent.
Model coverage
All measurements run against GPT-4o, Claude 3.5 Sonnet, Gemini 1.5 Pro, Perplexity Sonar Large, and Yi-Large (01.AI). Enterprise plans add custom models on request.
Refresh cadence
Pay-as-you-go: daily refresh. Monitoring: 6-hour refresh. Enterprise: hourly with on-demand recompute. Drift detection alerts within one cycle of statistically significant changes.
Mention extraction
Hybrid pipeline: regex for exact brand, NER for variants, and an LLM-based reranker for ambiguous mentions. Precision/recall measured quarterly against human-labeled gold set (current: 0.94 F1).
Sentiment & position scoring
Each mention is scored on sentiment (-1 to +1), position (1 to N within an answer), and descriptor list (adjectives associated with the brand). Aggregates compound to a single Visibility Index 0–100.
Statistical significance
We don't show changes below the noise floor. Day-over-day deltas require ≥2σ separation on a 14-day rolling baseline. Weekly reports include confidence intervals.
Data handling & retention
All prompt/response data encrypted in transit (TLS 1.3) and at rest (AES-256). Customer data segregated by tenant. 90-day retention default; configurable to 24 months on Enterprise.
Audit trail
Every Visibility Index data point traces back to the exact prompt, model, timestamp, and raw response. Enterprise customers get full export via API.
One last thing
If AI doesn't recommend you, your business is already invisible.
Find out where you stand in 30 seconds. Free. No credit card.