Methodology

wordradar publishes its scoring methodology so you can judge what the numbers mean before you bet a site on them. Every metric on this page is an estimate derived from public signals — none of it is official search-engine data.

Proxy-KD (SERP difficulty)

You feed the top-10 organic results for a keyword; the scorer classifies each result's domain and returns a 0-100 difficulty estimate with a per-result breakdown.

Base score: 50. Each classified result adjusts the total:

Clamping: the final score is clamped to 0-100; a table of ten major-brand shallow pages maxes out at 100 rather than 200.

One domain counts as one signal only, in priority order: free → UGC → brand → video → wiki (so yahoo.com counts as UGC, not brand). Suffix lists cover 18 free-hosting and 18 brand domains each.

Verdict bands: below 20 = very easy, 20-39 = winnable, 40-59 = moderate, 60+ = hard.

Honest caveat: these weights are initial values, not calibrated against a proprietary difficulty vendor. A free subdomain ranking top-10 is a strong weakness signal in practice, but treat the absolute number as a comparable heuristic — cross-check against the raw signal breakdown the API returns, not just the score.

First-seen dates (word ledger)

The ledger records the date a word first entered our suggest corpus — not the date it was first searched anywhere. In practice a word usually appears in suggest endpoints days-to-weeks before it shows meaningful volume, which is exactly the window the radar is for. But a first-seen date can also lag reality if our crawl missed early occurrences. Read it as "when our radar caught it", nothing stronger.

Word types are classified adaptively: a word seen in few periods is reported as untyped until enough evidence accumulates for a new-word / re-spike / evergreen call.

Trends risers

Rising queries come from our Trends collection pipeline with the fetch timestamp attached. Treat the list as a cached snapshot, not a live feed — check fetched_at in the response before comparing two pulls.

Data provenance

• Suggest corpora: multi-engine autocomplete expansion, deduplicated into a 1M+ word corpus
• Site radar: public "new/hosted site" mentions (HN etc.) with free-hosting detection
• Domain timeline: RDAP registration events cross-checked with Wayback Machine snapshots
• Stripe leaderboard: the public referral leaderboard, archived monthly

Corrections

If a number looks wrong, open an issue in the repository with the keyword and what you expected — methodology bugs get priority.