How a check works

We generate five buyer-intent prompts from the category, country and current UTC year:

1. Best {category} in {country}
2. Top {category} for small businesses in {country}
3. {category} alternatives for buyers in {country}
4. Which {category} do experts recommend in {country}?
5. Most trusted {category} {year} in {country}

Engine and parsing

The default is Gemini 3.5 Flash-Lite (Gemini 3.5 Flash as fallback), answering from model knowledge (no live web search). ChatGPT is not checked yet. Google Search grounding is optional and requires an eligible tier. Each report shows the actual engine, model and date; local mock reports are synthetic demonstrations. Names and domains are normalized with NFKC, lowercase and punctuation removal. A mention requires a whole normalized phrase match against the name, name without Inc/Ltd/LLC/Limited, or domain. Names are extracted heuristically from bold text, numbered/bulleted lists, headings, table first columns, links and names followed by domains. These can include false positives. Position is one plus the number of distinct extracted competitors appearing before your first mention; it is not a provider ranking.

Citations and score

We combine grounding citations and URLs found in answer text. Direct URLs supply their hostname. Google grounding redirect links are followed for at most three redirects after DNS safety checks at every hop. If resolution fails, a domain-shaped title can supply a labelled fallback; otherwise the domain is unresolved. Resolutions are cached for seven days. Domain matching strips www and uses a limited public-suffix list, so some suffixes may be misclassified.

For each answer: (mentioned ? 60 + 25 / position : 0) + (your domain cited ? 15 : 0). The score is the rounded mean across answers, from 0 to 100. A citation alone can contribute 15. A first-position mention with a citation scores 100.

Sample size and cadence

A free check is five answers, cached for 24 hours. Paid active prompts become due every seven days, one answer per prompt, subject to budgets and provider availability. Category prompts become eligible for refresh after 30 days, subject to daily caps and provider availability, and panels only publish with five distinct real prompts within 45 days and three named brands. Sample sizes and run ranges are shown.

Limits

Answers are non-deterministic. This small sample is not a census of AI recommendations. We do not claim accuracy of every extracted name, market share, endorsement, causation, sales uplift, comprehensive engine coverage or guaranteed rankings. Compare repeated samples and independently verify model claims.

When no answer in a report is grounded, the citation component does not apply: score = round(mean(mentioned ? 60 + 25 / position : 0) × 100 / 85). A first-position mention scores 100. If any answer is grounded, retain score = round(mean((mentioned ? 60 + 25 / position : 0) + (your domain cited ? 15 : 0))).