How the Intent Score works

What the Intent Score measures

The Intent Score measures how much of a travel creator's audience expresses real booking intent in public comments — asking prices, booking steps, itineraries — versus passive admiration. We classify every sampled comment, compute the share expressing intent, and rank creators against their cohort.

Intent Score and Intent Density are Tokopo's own analysis of public comments — they are not YouTube metrics.

The four buckets

  • BOOKING — explicit purchase or booking intent, or direct commercial questions. Pattern example: a comment asking the total trip cost and how to book the same package.
  • PLANNING — genuine trip research without purchase language. Pattern example: a comment asking whether the trip is feasible in five days on a mid-range budget.
  • APPRECIATION — praise and aspiration without action. Pattern example: a comment calling the destination a bucket-list dream, with no question attached.
  • NOISE — spam, self-promotion, off-topic. Pattern example: a sub-for-sub solicitation.

Examples are paraphrased patterns, not quotes — we never reproduce commenter text. Comments are analyzed in aggregate; commenter identity is stored only as a salted hash.

Sampling rules

  • Last 10 long-form videos per creator (Shorts excluded), each with at least 20 ingested top-level comments.
  • Up to 200 comments per video, in YouTube's relevance order. Relevance ordering is engagement-biased — the density we report is density among the comments YouTube surfaces, and we say so rather than pretend otherwise.
  • Minimum viable sample: 150 classified comments across 4+ videos. Below that, a creator gets no public score — "insufficient data" is rendered, never a guess.

The formula, with a worked example

W = B + 0.4·P, counting only classifications at ≥ 0.5 confidence. T = all classified comments including NOISE — a spam-heavy audience dilutes its own score. Density = (W + 50·p₀) / (T + 50), a Bayesian smoothing that prevents small samples from spiking; p₀ is the cohort mean (bootstrapped at 0.02 until 50+ creators are scored).

Worked example: 1,000 classified comments, 30 BOOKING and 50 PLANNING above confidence → W = 30 + 0.4·50 = 50. Density = (50 + 50·0.02) / (1,000 + 50) = 51/1,050 = 4.9%. The Tokopo Score is that density's percentile rank within the cohort at computation time.

Limitations

  • YouTube-only in v1. We measure the platform whose comments are officially accessible. Instagram intent measurement unlocks per-creator when a creator claims their profile and connects their account — by design, that data cannot be scraped.
  • Small-cohort caveat. A percentile over a small cohort is noisy. Until 25 creators are scored, we publish Intent Density (the absolute, stable number) and rank-out-of-n, and hold back the 0–100 Tokopo Score.
  • Relevance-order sampling bias, disclosed above.
  • Rank moves as the index grows. Density is comparable over time; the Score is relative to the cohort on the day it's computed.

Score versions

Any change to buckets, weights, smoothing, or sampling becomes a new score version. Historical scores are never overwritten. Current version: v1.0.

  • v1.0 — initial public methodology (2026): four buckets, W = B + 0.4·P, k = 50 smoothing, last-10-videos sampling.