How InfluencerUnion
estimates sponsorship rates

The InfluencerUnion creator index is built bottom-up from public data on the YouTube Data API v3. We do not buy third-party creator lists; every channel we list was surfaced by one of four primitives, all open-source and documented:

• Top-channel seed lists from Wikipedia ("List of most-subscribed YouTube channels" + regional and per-language variants).
• Trending sweeps via videos.list?chart=mostPopular across ~50 ISO regionCodes. Each region yields its top-50 currently trending videos; the channels behind those videos are enqueued for enrichment. 50 quota units / day.
• Niche-targeted search using search.list?type=channel with regionCode + relevanceLanguage biases and creator-discovery queries in the local language. Currently ~67 queries across 16 languages targeting finance, crypto, B2B, health, real-estate, beauty, fashion, fitness, parenting, education, automotive, home_diy, and tech.
• Comment-thread harvesting as a supplemental long-tail source. ~1% of comment-source channels are real creators, so we run this sparingly.

Every discovered channel is enriched with the public metadata YouTube exposes: title, description, country, default language, subscriber count, total view count, video count. We never request or store private data and we do not use OAuth scopes.

Every estimate on the site is the output of the same transparent formula. There are three modes, depending on how much community-reported data we have for that creator:

1. Pure algorithmic estimate · zero community data. Uses the formula below.
2. Partially calibrated · 1–4 community data points. We blend the algorithmic estimate with the reported percentile in proportion to the sample size.
3. Calibrated · 5+ community data points. We use the p20, p50, p80 of the reported values directly.

The pure-algorithmic mid-point formula:

estimate_usd = avg_views_30d × cpm[niche] / 1000 × multiplier[platform] × multiplier[format]

When avg_views_30d is missing, we fall back to a subscriber-based view-through ratio that decays with audience size (50% under 10K subs, 35% to 100K, 20% to 1M, 12% above). For follower-driven formats (post, story, tweet, reel) we replace the view term with subscriberCount / 1000 × 8 USD/k.

Low/high bounds widen the band by ±50% around the mid point for pure algorithmic estimates, and tighten to actual p20/p80 of reported deals once calibration kicks in.

CPM is the lever that captures niche-level brand spend per impression. We calibrated these against published 2024-2025 sponsorship rate cards from the YouTube creator-economy press and the Influencer Marketing Hub annual report. Default fallback is $12 when we can't classify the niche.

Platform multipliers reflect brand willingness-to-pay per impression on each platform. YouTube is the index (1.00x); Instagram is the most expensive per-view (mostly because Reel impressions are short and brand-tagged); TikTok is cheapest because its view counts inflate vs. retention.

Format multipliers reflect the production cost and brand real-estate of each deliverable. A dedicated video commands twice an integrated mention; a tweet is roughly a third.

Not every creator at every audience size is worth indexing. We use category-specific subscriber thresholds: high-value B2B-adjacent niches (finance, crypto, B2B, health, real-estate) admit creators from 10,000 subscribers, while saturated mass-market niches (music, comedy, entertainment) require 500,000 subscribers. The full thresholds are listed in the niche CPM table above.

This is a deliberate quality choice: a 10K-subscriber finance YouTuber is genuinely useful to brands; a 10K-subscriber music channel is not.

When a creator hasn't been classified yet (low confidence from the AI classifier), we apply the 10,000 holding threshold and re-run classification later with richer signal (recent video titles, Tier-2 reclassifier).

Every creator's primary niche is assigned by an AI classifier running on Gemini 2.5 Flash (primary) with Claude Haiku as fallback. The input is the channel snippet (title, description, country, language, subscriber count); for low-confidence cases we re-run with up to 20 recent video titles attached (Tier-2 path).

The classifier returns a single primary category from a fixed set of 26, up to 3 optional secondary categories, and a confidence score 0-1. We only persist labels with confidence ≥ 0.7; lower-confidence rows are held with a null category until the next reclassification pass.

Snippet-only high-confidence rate: ~80% of creators across our index.
Tier-2 high-confidence rate (with video titles): ~93% on the previously-ambiguous holding bucket.

• Discovery: daily — trending sweep + queue drain via the production discover daily command.
• Re-sync of indexed creators: weekly for top 10K by subscriber count, monthly for the rest. Subscriber counts on creator profile pages are accurate as of the most recent successful sync, displayed in the page header.
• Niche reclassification: quarterly, or sooner if a creator's recent uploads suggest a pivot.
• SEO content regeneration: every 90 days, or sooner if a creator's subscriber count changes by >25%.
• Pricing multipliers: we revisit the CPM and multiplier tables annually against fresh industry data. Last revision: 2026-Q1.

• Public channel metadata: channels.list, videos.list, and playlistItems.list on the YouTube Data API v3. We never request OAuth scopes and never see private user data.
• Community pricing submissions: anonymous deal-price submissions from creators and brands, used only as calibration input to the pricing model. Individual submissions are not displayed publicly.
• External seeds: Wikipedia top-creator list articles, parsed for YouTube external links. No scraping of Social Blade, HypeAuditor, or other paid aggregators.

• Estimates, not contracts. Every dollar figure on this site is an algorithmic estimate. Real deal prices vary substantially based on usage rights, exclusivity windows, content quality requirements, and creator-brand history.
• YouTube-first coverage. The current index is YouTube-only. TikTok, Instagram, X and Bilibili are supported in the pricing model and schema but ingestion workers for those platforms ship in later phases.
• Hidden-subscriber-count channels. ~1-3% of large channels (YouTube Originals, some media brands) hide their subscriber count. These get a best-effort estimate from view count and video output, and a tag noting the limitation.
• Language coverage skew. Our niche- targeted discovery currently runs in 16 languages. Creators publishing in Bengali, Tamil, Telugu, Egyptian Arabic, Swahili, and most Sub-Saharan African languages are under-represented relative to their actual creator- economy footprint. Expanding these is a 2026-Q2 priority.
• Recency of calibration. CPM and multiplier values are calibrated against publicly- published industry data; they lag the market by 6-12 months. We do not have access to private rate cards.
• No view-quality adjustment yet. The current model treats all views equally. Sponsored- content view-through rate vs. viral-clip retention is not factored in. This is a known weakness; it tends to overstate prices for creators with heavy short-form traffic.