Methodology
How AnswerLens measures AI visibility — in the open. Sophisticated SEOs distrust black-box scores, so every number here is explainable and every limit is disclosed.
What we query
Quick audits use 2 prompts, run once across Perplexity, Gemini, and Google AI Overviews. Saved-client audits use the five-engine methodology below. Each report lists its actual coverage, unavailable engines, and whether the data is live or a demo.
We generate 3 buyer-intent prompts for the brand's industry (best-of, comparison, alternatives, problem, and brand questions), then run each one 1× on every engine below. API answers approximate — but don't perfectly equal — what a logged-in consumer sees (different system prompts and personalization).
Google AI Overviews are retrieved through a licensed SERP provider — we never scrape Google. If an engine is unavailable for a query, we show that honestly rather than faking coverage.
Handling AI answer variability
AI answers are non-deterministic, so we never present a single run as truth. Each prompt is run 1× per engine and reported as Always (3/3), Sometimes (1–2/3), or Never (0/3). Treat trends, not single points, as signal.
The AI Visibility Score
A 0–100 composite of seven components, each weighted and aggregated with prompt-intent weighting (buyer-intent prompts count more):
| Component | Weight | How it's calculated |
|---|---|---|
| Mention rate | 30% | Share of prompt-runs where your brand appears at all. |
| Share of voice | 20% | Your mentions ÷ (your + all competitors') mentions. |
| Position | 15% | How early you're named in the answer (first beats buried). |
| Recommendation rate | 15% | Runs where AI actively recommends you, not just names you. |
| Citation rate | 10% | Runs (with any citation) that cite your own domain. |
| Sentiment | 5% | Positive + ½·neutral mentions ÷ total mentions. |
| Platform coverage | 5% | Engines where you appear ÷ engines monitored. |
Score bands
Detection
Brand and competitor mentions use case-insensitive, boundary-aware matching with alias and spacing tolerance (so "Hub Spot" matches "HubSpot") without false positives. Citations are resolved to a registrable domain and tagged as your domain, a competitor's, or third-party. Recommendation strength and sentiment are inferred from the language around each mention.
Demo vs live data
This build runs in demo mode by default: deterministic, realistic sample answers so you can explore the full workflow with zero API keys and zero cost. Add provider API keys and set AUDIT_MODE=live to query the real engines — each audit then reflects live AI answers. Every result is labeled with the data mode it used.