How scoring works

1. The question corpus

Every category holds 15 buyer intent questionswritten the way real customers actually ask: symptoms, goals, budgets, and trust concerns. Questions never name a brand, so every brand mention in an answer is the engine's own unprompted recommendation. Nothing is seeded and nothing is steered. Each question carries one of six intent types:

• discoveryNeed based, like: I cannot fall asleep at night, what actually helps?
• qualifiedConstraint specific, like: best option for women over 40, or which take insurance?
• comparisonChoice making, like: magnesium glycinate vs citrate, which should I buy?
• validationTrust checks, like: are online GLP-1 clinics legit and safe?
• pricingCost questions, like: what does treatment cost per month without insurance?
• switchingAlternative seeking, like: what should I try if melatonin does not work?

The full question set for each category is published on its category page, so brands always know exactly what is being asked on their behalf.

2. Daily rotation

The 15 questions are split into five sets of three. One set runs each day in round robin rotation: set 1 on day one, set 2 on day two, and set 1 again on day six. The full corpus is covered every five days while engine load stays minimal and steady. The three questions running today are highlighted on every category page. Comparing the same weekday five days apart always compares identical questions.

3. Collection

The day's set runs against ChatGPT, Claude, Perplexity, Google AI every day at 7:00 AM Eastern through their official APIs. Engines are asked cold: no system steering toward any brand, no follow ups, one question per call. We store the full answer text with its exact collection timestamp and, where the engine provides them, the sources it cited. Every collected answer is published verbatim on its own page as proof, with formatting preserved exactly as the engine wrote it, so any score can be audited back to the actual responses behind it.

4. Open brand discovery

Every answer is parsed for every brand the engine cites, not just a curated list. A brand we have never tracked before is added to the index the day an engine first recommends it, flagged as a new entrant, and stamped with its first seen date. Brands the engines stop citing derank naturally. We do not decide who ranks; the engines do.

To keep the tables brand versus brand, extraction excludes generic ingredients and compounds (magnesium, melatonin, semaglutide), prescription medication product names (Ozempic, Wegovy), retail marketplaces mentioned only as places to buy (Amazon, Costco), publishers and review sites (Healthline, Reddit), and government or medical institutions (FDA, Mayo Clinic).

5. Sentiment

Each mention records how that specific answer treats the brand: positive (recommended or praised), neutral (listed without judgment), or negative (warned against or criticized). Sentiment is assessed per answer at parse time, never edited, and rolls up on each brand page so you can see not just how often AI mentions a brand but how it talks about it.

6. Scoring and ranking

The core metric is AI visibility: the percentage of a category's answers on a given day that mention the brand. A brand at 80% visibility appeared in four out of five of that day's answers.

• Rankis computed daily from brands actually cited that day, ordered by visibility. Not being cited means falling off the board, not holding yesterday's spot.
• Average position tracks where in the answer a brand appears: first mention carries more buyer attention than fifth.
• Per engine breakdowns show where a brand is strong and where it is invisible across ChatGPT, Claude, Perplexity, Google AI.
• Day over day movement (places gained or lost, debuts, new entrant badges) is shown on every leaderboard.

7. The permanent public record

Every category and brand page carries a public rank history. Brand pages are never deleted: a brand that stops being cited keeps its page, its peak, its last cited date, and a day by day record of the decline. If it recovers, that shows here the same day. The record is append only; we do not edit history for anyone, including brands that work with us.

8. Sources and citations

Where engines cite their sources (Perplexity always, others when available), we store every URL and resolve it to its domain. Category pages show which sites drive recommendations in that market; brand pages show which sources appear around a specific brand; every answer page lists the exact URLs behind that answer. This is where AI visibility is won, and it is the foundation of the AI Visibility Audit.

Limitations

• API responses can differ from what consumer apps show, since apps layer on search, memory, and personalization. Treat scores as directional.
• Consecutive days run different question sets, so day over day movement can reflect the question mix as well as engine behavior. Identical questions recur every five days; those comparisons are exact.
• Brand extraction and sentiment are automated judgments made at parse time. Canonical names are matched case insensitively; sub brands may be attributed to the parent brand.
• Newly discovered brands start with the engine's own naming and the best available website match; profiles fill in over time.
• One answer per question per engine per day is a sample, not a census. Small categories move more than large ones.

A note on health

WellRank reports what AI engines say, including when they are wrong. Nothing here is medical advice. AI engines can and do hallucinate ingredients, dosages, prices, and clinical claims, which is exactly why brands in regulated health categories should be monitoring this. If you spot an AI answer making a false claim about your brand, we want to hear about it.