01
Start with buyer questions
We read the public homepage, respecting its robots policy, and identify your brand and category. The free scan asks a few questions (currently three) in every available assistant. The full audit uses the number of questions shown at checkout across the available providers. An AI model writes the questions in plain, everyday wording from only your category, market and language — it never sees your brand, website or competitors, so the questions cannot steer answers toward you. Each one asks the assistant to name, recommend or compare options. Your project keeps the same questions, so paid audits and monitoring repeat them and stay comparable over time; if writing them fails, fixed neutral templates are used instead. Choose a target market and question language independently. The report interface remains English.
02
Preserve the evidence
We store the question, provider, model, timestamp, raw answer, citations, token usage and estimated costs. A separate structured analysis extracts commercial brand mentions, recommendation strength, sentiment and supporting quotes. Quotes and brand occurrences are checked against the original text. Automated extraction can still make mistakes; inspect the original answers. Direct competitors are distinguished from tools, publishers and uncertain entities. Only the target and direct competitors contribute to competitive share of voice and mention order.
03
Calculate the score
| COMPONENT | WEIGHT | MEASUREMENT |
|---|
| Mention rate | 25% | Successful answers containing the target brand. |
| Recommendation rate | 20% | Successful answers positively suggesting the target brand as a choice. |
| Mention order | 5% | Reciprocal first-mention order among extracted commercial brands, averaged across successful answers. |
| Share of voice | 20% | Target brand mentions divided by all distinct brand mentions. Each brand is counted once per answer. |
| Citation rate | 15% | Successful answers that cite the target website or its subdomains. |
| Cross-model presence | 10% | Providers mentioning the brand at least once divided by providers with successful answers. A one-provider scan offers no evidence about other models. |
| Sentiment | 5% | Positive target mentions score 1, neutral 0.5 and negative 0, averaged across all successful answers. |
The final score is the weighted average of these components on a 0–100 scale. We normalize by the sum of configured weights. Every report preserves the weights used at generation time. Failed answers are excluded and reported separately; an entirely failed audit has no score. With fewer than two successful providers, cross-model presence is unavailable and its weight is excluded before normalization.
04
Review competitors and cited sources
Competitors are commercial alternatives explicitly named in answers. Source intelligence aggregates provider-returned citations. A source marked for review was cited in an answer naming other brands. We do not check the cited pages for brand coverage. Co-occurrence does not prove the source endorses a competitor or caused an AI recommendation. “Sources AI trusts” is shorthand for observed citations, not an independently measured trust score.
05
Turn patterns into next steps
Recommendations connect observed gaps to concrete pages, product information, structured data or third-party opportunities. Homepage signals such as a missing pricing link are indicative, not a full technical crawl. Impact and difficulty are qualitative priorities, not predicted traffic or revenue gains.
Reading changes over time
Answers can change without any changes to your website. Compare the same questions, markets and providers. A free scan is a small sample; a single answer can move its mention rate by ten percentage points or more. Monitoring provides repeated snapshots, not guaranteed placement. Plan budgets may pause further checks until capacity is available.
