How PAI Scores Work

PAI measures public attention and visibility — how much the world is currently watching, reading, and talking about a person. It does not measure talent, achievement, or moral worth.

Important: Scores reflect real-time data signals and may fluctuate daily based on news cycles, viral moments, and Wikipedia edit activity. A high score means more people are paying attention right now — nothing more.

Two Scores

Popularity Score

Overall public mindshare. Combines search interest, Wikipedia traffic, news coverage, social reach, conversation volume, and Wikipedia presence across languages.

Heat Score

Momentum right now. Measures spikes in Wikipedia pageviews, search interest, news velocity, and social conversation compared to recent baselines.

Sentiment is tracked separately and shown on person profiles. It is never used to calculate Popularity or Heat.

Popularity Components

Component	Weight	Source	Status
Search interest	15%	Google Trends proxy	Mock
Wikipedia attention	15%	Wikimedia Pageviews API	Live
News / media coverage	25%	News API proxy	Mock
Social reach	15%	Social data proxy	Mock
Conversation volume	15%	Conversation proxy	Mock
Enduring prominence	15%	Wikidata (sitelinks)	Live

Heat Components

Component	Weight	Calculation
Search spike	30%	log₁₀(7d avg / 90d avg) / log₁₀(50) × 100
Pageview spike	25%	log₁₀(7d avg / 90d avg) / log₁₀(50) × 100
News velocity	25%	Articles this week vs. baseline
Social velocity	20%	Conversation velocity ratio

Missing Data Policy

When a signal is unavailable, it is excluded from the denominator of the weighted average — it never pulls the score down. A person with 3 live signals scores as if those 3 signals are the full picture, rescaled to 100%.

Coverage score shows what fraction of signals were available. Scores with low coverage should be interpreted with caution.

Coverage label	Threshold	Meaning
High coverage	≥ 70% signals	Score is well-supported by available data
Partial coverage	40–69% signals	Score is directionally correct but some signals are missing
Insufficient data	< 40% signals	Too few signals to form a reliable score

Score Model

All scores are tagged with a model version (currently v1). When the scoring formula changes, scores from different versions will not be compared to each other.