YouTube Influencer Audit — Fake Subscribers, Engagement & Sponsors

Vet a YouTube creator before you sponsor them. Give channels and get the agency due-diligence brief for each:

• Engagement benchmarked to the subscriber tier, with long-form and Shorts analysed separately.
• Fake-subscriber and bought-view signals.
• Reach health and momentum.
• Which brands recently sponsored them.
• Brand-safety flags.
• A CPM rate card for long-form and Shorts.
• Authenticity, reach-health and influence scores, with a verdict and a one-line summary.

It reads public channel data directly. No API key and no extra scraper to pay for. Agent-callable.

Built for influencer-marketing agencies, brand partnership and sponsorship teams, and creator marketplaces.

Tiers (the deep layer runs only when you ask)

• Channel audit (default, a few seconds): subscriber-tier benchmarks, bought-subscriber / bought-view tells, reach health + momentum, upload cadence, sponsor share + recent sponsors, brand safety, rate card, three scores, verdict.

• Comment forensics (deepComments: true): reads the newest comments on recent uploads, where spam bots live and top-comment sorting hides them. It grades:

  • Telegram/WhatsApp/"investment" scam bots
  • generic/emoji-only comments
  • duplicates
  • commenters still on YouTube's auto-generated @user-… handle
  • script mismatch

  The result is a commentAuthenticityScore blended into the verdict.

These are authenticity signals from public channel data for due diligence. They are not an accusation that a creator is fraudulent. Benchmarks and the rate card are indicative. Verify before contracting.

What makes it deep

Dimension	What it computes
Long-form vs Shorts	Shorts count views differently and would distort every rate, so they're analysed separately. Each format gets its own median views, like rate and rate card, and the audit flags when reach really comes from Shorts.
Do subscribers watch?	Median views ÷ subscribers vs the tier norm. Subscribers who never watch are the classic bought-subscriber pattern.
Lifetime views per subscriber	Organic channels earn 100+ lifetime views per subscriber. A large subscriber base on very few views points to bought subscribers.
Engagement sanity	Like rate (views without likes → bought views), comments per 1k likes (lopsided → bought likes), view-count uniformity.
Reach health + momentum	Recent views per upload vs the channel's own lifetime average (rising / steady / cooling / declining), plus upload activity. Price on current reach.
Sponsors	Share of uploads with sponsor or promo-code disclosures, plus the brands that recently sponsored the creator (competitor intel).
Credibility gating	Verified/established channels aren't flagged as fake for naturally low engagement. Their dormant subscribers show up as low reach health instead.
Commercials	CPM rate card on median views × engagement quality, long-form and Shorts.
Brand safety & fit	Risky-content flags (adult / gambling / crypto-promo / engagement-bait / profanity), YouTube's own not-family-safe flag, optional brand-keyword fit.
Verdict	authenticityScore + reachHealthScore + influenceScore + signal + summary + findings.

Verdicts: authentic-healthy-engagement · authentic-modest-engagement · inactive-audience-low-reach · inflated-subscribers-suspected · bought-engagement-suspected · engagement-authenticity-concerns · authenticity-concerns--review · brand-safety-review · subscriber-count-hidden · not-found.

Why not just ask an AI agent?

An agent can open a channel page. It can't, in one call and for a whole roster:

• pull exact views and likes for the latest uploads,
• split Shorts from long-form,
• read comment counts and the newest comments,
• extract who sponsored the creator,
• benchmark against the right tier,
• run the bought-subscriber forensics,
• roll it all into consistent scores.

Input

Field	Type	Required	Description
channels	array of strings	✅*	@handles, channel URLs or UC… ids.
brandKeywords	array of strings		Topics for a brand-fit read.
datasetId	dataset	✅*	…or read channels from another dataset (e.g. a YouTube channel-finder run).
deepComments	boolean		Comment forensics add-on (default off).
commentVideosSampled	integer		Uploads sampled for comments (default 3).
maxChannels	integer		Cap per run (default 20).
webhookUrl	string		POST the audits here as JSON.

* provide channels or datasetId.

Example input

{ "channels": ["@mkbhd", "@veritasium", "@MrBeast"], "brandKeywords": ["tech", "science"] }

Output

One audit per channel. Example (real run, trimmed):

{
  "handle": "@mkbhd",
  "title": "Marques Brownlee",
  "verified": true,
  "tier": "celebrity",
  "subscribersCount": 21400000,
  "analysedFormat": "long-form",
  "medianViews": 8852726,
  "viewsToSubscribersRatio": 0.414,
  "likeRatePct": 1.24,
  "commentsPer1kLikes": 69.1,
  "lifetimeViewsPerSubscriber": 270.3,
  "momentum": {"label": "rising", "recentVsLifetimeViewsRatio": 3.3},
  "cadence": {"avgDaysBetweenLongForm": 3.6, "shortsShareOfUploads": 13},
  "shorts": {"count": 2, "medianViews": 7665950, "likeRatePct": 2.37},
  "recentSponsors": [],
  "authenticityScore": 94, "reachHealthScore": 83, "influenceScore": 82,
  "estimatedRateCardUsd": {"low": 79675, "mid": 132791, "high": 212465, "currency": "USD"},
  "signal": "authentic-healthy-engagement",
  "summary": "@mkbhd - celebrity tier (21,400,000 subscribers) - median 8,852,726 long-form views (41% of subs) - like rate 1.2% - reach rising - authenticity 94, reach health 83, influence 82 -> authentic-healthy-engagement"
}

On a channel with sponsor disclosures, recentSponsors lists the brands (e.g. a real run on @MrBeast returned ["Krispy Kreme", "UPLIFT Desk", …]). An unverified channel whose subscribers far out-number its views returns inflated-subscribers-suspected.

Use cases

• Sponsorship vetting: confirm the audience is real and watching now before you pay.
• Roster screening: audit a shortlist or a channel-finder dataset in one run and rank by authenticity and reach health.
• Rate negotiation: an independent, views-based fair price for long-form and Shorts integrations.
• Competitor intel: see which brands sponsor the creators in your niche.
• Brand safety: risky content, not-family-safe channels, ad-fatigue.
• AI agents: a structured verdict + one-line summary to gate outreach.

One API call, JSON back

Run it synchronously: one POST starts the run and returns the results in the same response. No run to poll, which is what an AI agent wants mid-task:

curl -X POST "https://api.apify.com/v2/acts/centorprat~youtube-influencer-audit/run-sync-get-dataset-items" -H "Authorization: Bearer <APIFY_TOKEN>" \
     -H "Content-Type: application/json" -d '{"channels": ["@mkbhd", "@veritasium"]}'

The body takes the same fields as the normal input. Same pay-per-event price; failed checks aren't charged.

For AI agents / MCP

https://mcp.apify.com/?tools=actors,docs,centorprat/youtube-influencer-audit

Loads as centorprat--youtube-influencer-audit. Claude Code:

claude mcp add apify -- npx -y @apify/actors-mcp-server --actors centorprat/youtube-influencer-audit

FAQ

Do I need a YouTube API key? No. It reads public channel pages, the channel's uploads feed and public video pages.

Why separate Shorts? Shorts views are counted differently and run far higher per subscriber. Mixing them hides how a long-form integration will actually perform.

How many uploads are analysed? The channel's latest 15 uploads, with comment counts read for the most recent long-form ones. Videos under 48 h old are left out of the rates.

What's the difference between authenticity and reach health? Authenticity asks whether the subscribers and engagement are real. Reach health asks whether subscribers watch now. A channel that peaked years ago can be authentic and still a poor buy today.

Does it accuse creators of fraud? No. It surfaces signals for your review, and verified/established channels are gated against false flags.

Hidden subscriber counts? Reported as subscriber-count-hidden. Views-based metrics still run; subscriber benchmarks are skipped.

Does it profile commenters? No. Comment forensics is aggregate (percentages only). No commenter is named, and nothing is inferred about nationality.

How fast? A few seconds per channel, a bit more with comment forensics. Channels run in parallel.

Cost? Pay-per-event: one event per channel audited, plus one per channel when comment forensics runs. There's no underlying scraper bill.

Data & attribution

Public YouTube channel and video data. These are signals for influencer due diligence, not accusations or guarantees. Rate-card and benchmark figures are indicative.

Related actors

• TikTok Influencer Audit
• Instagram Influencer Audit