Fake Engagement & Bot Account Detector — Instagram, TikTok & YouTube Audience Authenticity API

Score any creator's audience for fake followers and bot engagement across Instagram, TikTok and YouTube in a single call. Give it a list of handles; it returns typed JSON rows — authenticity score, bot likelihood, verdict, a weighted signal breakdown, and the raw follower/engagement metrics behind every number, so you can re-derive the maths yourself. Schema below.

Built for creator platforms, influencer agencies and brand-safety pipelines that check audiences at volume: priced per profile analyzed, not per report.

Why agents use this actor

• Deterministic typed output — one row per profile, stable field names, every nullable field declared in the dataset schema.
• Per-profile pricing — one profile-analyzed event per scored profile, so an autonomous budget maps 1:1 to work done. Profiles whose engagement could not be retrieved are not charged.
• Never guesses — if engagement data is unavailable, verdict is null, confidence is insufficient-data, and unavailableSignals names exactly which signals were missing. Nulls are never scored as zeros.
• Explainable, not a black box — every signal ships with its value, status, weight and a plain-language explanation an agent can quote straight to a user.
• No auth, no cookies — public data only. Retries with backoff on 429/5xx, 30 s timeouts, per-item pushes.
• Three platforms, one contract — the same row shape for Instagram, TikTok and YouTube.

Input schema

Field	Type	Required	Default	Description
profiles	string[]	no*	[]	"instagram:handle", "tiktok:handle", "youtube:@handle", or a plain profile URL (platform auto-detected).
instagramUsernames	string[]	no*	[]	Instagram usernames, with or without @.
tiktokUsernames	string[]	no*	[]	TikTok usernames, with or without @.
youtubeChannels	string[]	no*	[]	YouTube handles (@MrBeast) or channel IDs (UC...).
maxPostsPerProfile	integer	no	12	Recent posts/videos sampled per profile for engagement, variance and cadence. 1–50.
proxyConfiguration	object	no	{ "useApifyProxy": true }	Apify Proxy settings. TikTok and YouTube run on datacenter/auto; Instagram automatically uses the residential group, billed at the standard residential rate.

* At least one profile across the four lists is required. All four lists are merged and de-duplicated.

Output schema

One row per profile.

Field	Type	Nullable	Description
platform	string	no	instagram, tiktok or youtube.
handle	string	no	Canonical handle as published by the platform.
profileUrl	string	no	Public profile URL.
displayName	string	yes	Display name / full name / channel title.
verified	boolean	yes	Verification badge; null where the platform hides it.
isPrivate	boolean	yes	True for private accounts.
followers	integer	yes	Followers / subscribers.
following	integer	yes	Accounts followed. Null on YouTube — it does not publish this.
posts	integer	yes	Total posts / videos.
totalLikes	integer	yes	Lifetime likes across all videos (TikTok).
avgLikes	integer	yes	Average likes per sampled post (TikTok: lifetime likes ÷ video count).
avgComments	integer	yes	Average comments per sampled post (Instagram).
avgViews	integer	yes	Average views per sampled video (TikTok, YouTube, Instagram video posts).
engagementRate	number	yes	Fraction of audience engaging — 0.032 = 3.2%.
benchmarkEngagementRate	number	yes	Typical rate for this platform at this audience size — the benchmark used.
engagementBasis	string	yes	Which observable metric fed the rate: recent-posts-likes-comments, lifetime-avg-likes-per-video, recent-video-views-per-subscriber.
postsAnalyzed	integer	yes	Posts actually sampled.
authenticityScore	integer	yes	0–100. Null when engagement could not be retrieved.
botLikelihood	number	yes	0–1 weighted share of signals pointing to inauthentic engagement.
verdict	string	yes	authentic (≥80) / likely-authentic (65–79) / suspicious (45–64) / likely-fake (<45). Null when confidence is insufficient-data.
confidence	string	no	high / medium / low / insufficient-data, from how many signals were retrievable.
signals	object[]	no	{ name, value, status, weight, explanation } per signal. status ∈ ok/warn/bad/unavailable.
redFlags	string[]	no	Explanations of every signal that scored bad.
greenFlags	string[]	no	Explanations of every healthy signal.
unavailableSignals	string[]	no	Signals with no retrievable data for this profile.
bio, externalUrl, avatarUrl	string	yes	Profile completeness inputs, returned raw.
channelId	string	yes	YouTube channel ID.
analyzedAt	string	no	ISO 8601 timestamp.
error	string	yes	Always null on pushed rows — failures go to the SUMMARY record.

The signals

Signal	Weight	What it catches
engagementRate	3	Engagement graded against the typical rate for that audience size, never an absolute — large accounts naturally engage lower. Far below tier reads as inactive/purchased followers; implausibly above reads as purchased engagement.
likeToCommentRatio	2	Bot engagement is likes-only. Real audiences comment.
engagementVariance	2	Organic reach is spiky. Near-identical numbers post to post are a delivered quota.
followerFollowingRatio	1	Mass-follow and follow/unfollow growth tactics.
postingCadence	1	Burst-and-dormancy publishing patterns.
accountCompleteness	1	Bio, link, avatar, name and verification — disposable accounts skip them.

The score is the weighted share of available signals that came back clean. Unavailable signals are excluded from both numerator and denominator and lower confidence instead.

Example rows (real output)

{
  "platform": "instagram", "handle": "nasa",
  "profileUrl": "https://www.instagram.com/nasa/", "displayName": "NASA",
  "verified": true, "isPrivate": false,
  "followers": 104398251, "following": 92, "posts": 4882,
  "avgLikes": 457668, "avgComments": 1488, "avgViews": 1514238,
  "engagementRate": 0.004398, "benchmarkEngagementRate": 0.015,
  "engagementBasis": "recent-posts-likes-comments", "postsAnalyzed": 12,
  "authenticityScore": 75, "botLikelihood": 0.25,
  "verdict": "likely-authentic", "confidence": "high",
  "signals": [
    { "name": "engagementRate", "value": 0.004398, "status": "warn", "weight": 3,
      "explanation": "0.44% engagement — 0.29x the 1.5% typical for this audience size. Below the healthy band for the tier." },
    { "name": "engagementVariance", "value": 1.259, "status": "ok", "weight": 2,
      "explanation": "Engagement varies by 126% across recent posts — the spiky pattern of organic reach." }
  ],
  "redFlags": [], "greenFlags": ["..."], "unavailableSignals": [],
  "analyzedAt": "2026-08-15T10:09:49.079Z", "error": null
}

{
  "platform": "tiktok", "handle": "charlidamelio",
  "followers": 159343052, "following": 1433, "posts": 3221,
  "totalLikes": 12306793939, "avgLikes": 3820800, "avgComments": null, "avgViews": 2964850,
  "engagementRate": 0.023978, "benchmarkEngagementRate": 0.025,
  "engagementBasis": "lifetime-avg-likes-per-video", "postsAnalyzed": 10,
  "authenticityScore": 100, "botLikelihood": 0, "verdict": "authentic", "confidence": "medium",
  "unavailableSignals": ["likeToCommentRatio", "postingCadence"]
}

{
  "platform": "youtube", "handle": "@MrBeast",
  "channelId": "UCX6OQ3DkcsbYNE6H8uQQuVA",
  "followers": 513000000, "following": null, "posts": 997,
  "avgViews": 110166667, "engagementRate": 0.21475, "benchmarkEngagementRate": 0.1,
  "engagementBasis": "recent-video-views-per-subscriber", "postsAnalyzed": 12,
  "authenticityScore": 100, "botLikelihood": 0, "verdict": "authentic", "confidence": "medium",
  "unavailableSignals": ["followerFollowingRatio", "likeToCommentRatio"]
}

Error semantics

• Bad input — no valid profile in any list: run fails fast, exit code 1, status message naming the accepted formats. Nothing is charged beyond actor-start.
• Per-profile failure — extraction errors never abort the run and never appear in the paid dataset. They are collected in the SUMMARY key-value record as { platform, handle, error }, so retry logic can pick them up: GET /v2/key-value-stores/{storeId}/records/SUMMARY.
• Partial data — a profile that resolves but exposes no engagement is pushed with verdict: null, confidence: "insufficient-data" and populated unavailableSignals, and is not charged.
• Zero results — if every profile fails, the run exits 1 with the first error in the status message.
• SUMMARY always contains { requested, analyzed, insufficientData, failed, failures, finishedAt }.

Use from AI agents (MCP)

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com/?tools=apricot_blackberry/fake-engagement-detector",
      "headers": { "Authorization": "Bearer <YOUR_APIFY_TOKEN>" }
    }
  }
}

Works in Claude, Cursor, ChatGPT deep-research connectors and any MCP client; the input schema above is the tool's parameter schema.

Use from code

curl -X POST "https://api.apify.com/v2/acts/apricot_blackberry~fake-engagement-detector/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"profiles":["tiktok:charlidamelio","youtube:@MrBeast"],"maxPostsPerProfile":12}'

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('apricot_blackberry/fake-engagement-detector').call({
    profiles: ['tiktok:charlidamelio', 'youtube:@MrBeast'],
    maxPostsPerProfile: 12,
});
const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items.map((i) => [i.handle, i.verdict, i.authenticityScore]));

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("apricot_blackberry/fake-engagement-detector").call(run_input={
    "profiles": ["tiktok:charlidamelio", "youtube:@MrBeast"],
    "maxPostsPerProfile": 12,
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    print(item["handle"], item["verdict"], item["authenticityScore"])

Use from automation platforms

• n8n / Make / Zapier — pick the Apify integration, choose Run actor, select Fake Engagement & Bot Account Detector, paste the input JSON above.
• LangChain / LlamaIndex — wrap it with the Apify actor tool (ApifyActorsTool / ApifyWrapper) using actor id apricot_blackberry/fake-engagement-detector.
• Webhooks — Apify can fire a webhook on ACTOR.RUN.SUCCEEDED/FAILED for this actor, so a brand-safety pipeline picks up verdicts on run completion without polling.

Pricing

Pay per event:

Event	When it fires
actor-start	Once per run.
profile-analyzed	Once per profile that returns a real verdict. Profiles with confidence: "insufficient-data" and failed extractions are not charged.

Instagram profiles route through Apify residential proxy, billed to your account at the standard residential rate.

FAQ

How do you detect fake followers without follower-list access? By the fingerprint that purchased audiences leave in public metrics: engagement far below the norm for the audience size, likes without comments, and suspiciously uniform post-to-post numbers. Every one of those is computed from data the platform publishes on the profile itself.

Can I check Instagram, TikTok and YouTube in one run? Yes — that is the point. Mix platforms freely in profiles; each row states its own platform and engagementBasis.

What does a good engagement rate look like? It depends entirely on audience size, which is why every profile is graded against benchmarkEngagementRate for its tier rather than a fixed number. A 1.5% rate is healthy for a 10M-follower account and weak for a 5K one.

Why is avgComments null on TikTok and YouTube? Those platforms publish likes and views on the profile surface but not per-post comment counts. The like-to-comment signal is then reported as unavailable and excluded from the score — it is never assumed to be zero.

Is a likely-fake verdict proof of fraud? It is a statistical assessment of public engagement patterns, with the full evidence in signals and redFlags so you can review the reasoning behind it.

How many profiles can I check per run? As many as you pass. Profiles are processed sequentially with retries; a 100-profile run is routine.

Changelog

1.0 — Initial release. Instagram, TikTok and YouTube extraction; six-signal weighted authenticity scoring with tier-relative engagement benchmarks; typed dataset schema with per-signal breakdown; per-profile pay-per-event pricing.

---

中文说明（刷粉检测 / 假粉丝识别）

给达人打款之前，先查一查。 输入 Instagram / TikTok / YouTube 创作者账号，输出机器人粉丝/虚假互动评分，帮助你在投放前识别刷粉账号、僵尸粉和虚假互动。

• 覆盖三大平台：Instagram、TikTok、YouTube
• 评分为概率性参考指标，配合人工复核使用
• 按结果计费，批量筛选达人名单成本可控

适合 MCN、达人投放团队、跨境电商卖家在建联和签约前做尽调。Apify 每月 $5 免费额度，无需信用卡。

For AI agents

Bot-likelihood scoring rows with per-signal breakdown — machine-consumable for automated creator vetting pipelines:

$curl -X POST "https://api.apify.com/v2/acts/apricot_blackberry~fake-engagement-detector/run-sync-get-dataset-items?token=$APIFY_TOKEN" -H "Content-Type: application/json" -d '{"profiles": ["https://www.tiktok.com/@natgeo"]}'