Channel Attention Audit

The surface we map

A YouTube channel leaks attention across five off-platform surfaces: cross-platform repurposing, podcast and audio, community, the collaboration network, and search presence. YouTube's own analytics measure the views you already earn on YouTube. They say nothing about the audiences gathering on the platforms next door, the shows you could guest on, or the communities forming around your topic.

The Channel Attention Audit reads what a channel publicly signals about those five surfaces and scores each one. It is not a subscriber or watch-time report, and it is not an editing or thumbnail retainer. It maps where a channel's audience is reachable beyond the channel itself, which is the part no on-platform dashboard shows you.

The evidence standard

Every dimension score and every named leak must cite fetched channel data as a short verbatim datum paired with a source pointer (for example a video title, a channel description, or a count). Before the audit is shown to you, a mechanical gate checks each datum against the exact field it cites: a text datum must be a verbatim substring of that field, normalized only for case and whitespace, and a number must equal a fetched count exactly, or a duration must convert to seconds and match exactly.

A dimension whose evidence does not trace is marked insufficient-signal rather than scored on a guess. A leak whose evidence does not trace is dropped, not paraphrased into place. No metric is invented. This is the part a commodity content review does not do: the prompt instructs the standard, and a separate gate enforces it, because an instruction is not enforcement.

How it works

1. Fetch. We pull roughly 25 recent videos plus channel metadata (titles, descriptions, counts, and the links in descriptions) through the YouTube Data API. We do not read transcripts at this depth, so the audit reasons only from public metadata.
2. Score five dimensions. Each surface is scored 0 to 20 and banded, then combined into a 0 to 100 composite that is labeled an estimate. D1 reads cross-platform repurposing signals (shorts-to-long-form cadence, links to other platforms). D2 reads explicit podcast and audio signals only. D3 reads community signals (Discord, Skool, Circle, newsletter links). D4 reads collaboration signals in titles and descriptions. D5 reads linked web presence.
3. Name the leaks. The audit returns up to five substantiated leaks, each with a severity and each routed to the Drosjer service that closes it. A surface with no observable signal produces no leak.

Accuracy and limitations

• Shorts are classified by a duration heuristic, so borderline durations are ambiguous. We label this as a heuristic rather than a measured fact.
• The podcast and audio dimension is metadata-only. It moves on a positive observable signal (a named episode, a podcast link, guest phrasing) and never infers a podcast presence from its absence.
• The search dimension reads the linked web presence visible in descriptions. It does not query search engines and does not assert rankings or domain authority.
• The composite score is a model estimate. The gate drops untraced evidence and re-marks dimensions insufficient; it does not recompute or inflate the scores.

What we do not claim

We read only what a channel publicly signals. Where a surface shows no signal, the audit reports insufficient signal rather than declaring the surface definitively absent. We never claim a leak we cannot trace to fetched channel data, and we never present an estimate as a measurement.