Every number on this site comes from public, keyless APIs — no scraping of protected search surfaces, no paid data resold. The scoring formula is deterministic and published below. If you disagree with a score, you can compute it yourself.
| Source | What it tells us | Access |
|---|---|---|
| Wikimedia Pageviews | Public attention per topic — daily article views for every tracked entity | official API · no key |
| GDELT DOC 2.0 | Global news volume over time — how much the world's press writes about a topic | official API · no key |
| Google autocomplete | Live demand expansion — the phrases real people type, used for keyword discovery | public endpoint · polite rates |
| 6 more suggest engines | Per-platform keyword ideas in the suggest tool | public · cached 10 min |
No SERP scraping of Google results. No paid search-volume resellers at this stage — if we add volume enrichment later, we'll disclose it per metric.
Seasonality (day-of-week) is removed from every series before scoring. All inputs come from a 120-day daily history per topic.
growth = mean(last 14d) vs mean(prior 14d), after de-seasonalizing
momentum = OLS slope of last 14d, normalized by level
breakout = max robust z-score of last 7d vs prior 30d (MAD-based)
agreement = share of sources growing >10% (confidence)
fad = spike-and-decay shape detector (penalty -25)
score = 0.4*growth + 0.2*momentum + 0.2*breakout + 0.2*agreement
-> scaled to 0..100, minus the fad penalty| Level | Means |
|---|---|
| exploding | score ≥ 70 and growth > +25% — strong multi-signal acceleration |
| rising | score ≥ 55 — attention climbing faster than most tracked topics |
| steady | score ≥ 35 — active, real demand, no spike |
| cooling | growth < −15% — fading from a recent peak |
| flat | everything else |
Wikipedia attention and news volume are proxies for search demand — correlated, not identical. Autocomplete phrases show demand language, not volume. We don't fabricate absolute search numbers like some paid tools; we'd rather show fewer, honest signals. Data health for every source is logged in each weekly report.
The daily dataset is public JSON — no key, cached 5 minutes: GET /api/trends. Per-engine suggestions: GET /api/suggest?engine=google&q=… (120 req/min per IP). Deep long-tail mining: GET /api/mine?q=… (2 runs/min per IP, cached 24h). Explore trends in the trend grid.