Every formula, threshold, and scoring weight published in full. This is the exact methodology powering every PulseROI creator intelligence report and 24/7 monitoring alert.
Every creator analyzed by PulseROI receives a composite PulseAuthenticity Score (0-100). This score is computed as a weighted average of 8 independent mathematical heuristics, each designed to catch a specific type of YouTube fraud.
Evaluates how much of a channel's subscriber base actively watches new content. When subscribers are purchased from SMM panels, they never return to watch videos, creating a severe gap between subscriber count and actual viewership.
A channel with 1,000,000 subscribers averaging only 4,000 views per video has an SVR of 0.4%. Those subscribers are almost certainly fake or the channel has terminal audience decay. Real, organically-grown channels maintain 5-20% SVR.
Indian SMM panels sell 1,000 YouTube subscribers for as low as ₹40-150. These purchased subscribers use real SIM-card-authenticated mobile accounts, making them harder to detect than Western datacenter bots. SVR is the single most reliable signal for catching this pattern.
Measures the proportion of viewers who actively interact with content. Both abnormally low and abnormally high engagement rates are red flags — low indicates purchased views with no real audience, high indicates coordinated engagement pods or bot-like activity.
A video with 500,000 views but only 200 likes (ERA = 0.04%) almost certainly has purchased views. Conversely, a video with 10,000 views and 2,500 likes (ERA = 25%) is likely being artificially boosted by engagement pods or like-farms.
Indian WhatsApp engagement pods are a uniquely prevalent fraud pattern. Groups of 50-200 creators agree to immediately like and comment on each other's new videos. This produces ERA values of 15-25% that look impressive but represent zero genuine audience interest.
Bots and automated scripts can easily click 'like' buttons at scale, but writing contextually relevant comments requires far more sophistication. A massive gap between likes and comments is a strong signal of automated like-farming.
If a video has 50,000 likes but only 120 comments, the LCR is 416:1. Real audiences don't behave this way — they comment at roughly 1 comment per 15-50 likes. An LCR above 100:1 is a strong bot signal.
Indian creator WhatsApp groups specifically coordinate comment exchanges. When LCR drops below 5:1, it typically indicates a reciprocal commenting ring where members write generic praise ('mast video bhai', 'great content') on each other's uploads.
Measures how many viewers convert into positive reactions. On YouTube, roughly 2-8% of genuine viewers click like. Ratios far outside this range indicate either purchased views (low LVR) or purchased likes (high LVR).
When views are purchased but likes are not, LVR collapses below 0.5%. When likes are purchased but views are organic, LVR exceeds 15%. Both patterns are mathematically detectable and indicate distinct types of fraud.
Indian view-purchase services typically deliver views without corresponding likes or comments. This creates a distinctive LVR signature that is nearly impossible to mask without purchasing proportional likes as well — which doubles the fraud cost and is therefore less common.
Comments require the highest effort from viewers — they must watch, form an opinion, and type a response. A channel with millions of views but almost no comments has an audience that doesn't genuinely engage.
CVR below 0.02% on a video with 100K+ views strongly indicates view farming. CVR above 1% with low quality comments may indicate comment pod activity or controversy-driven engagement.
The CVR threshold is particularly important for Indian regional language channels where comment culture varies significantly. Tamil and Telugu audiences comment more actively than Hindi audiences on average, which PulseROI adjusts for using our language detection engine.
Real human audiences are inherently unpredictable. Some videos go viral, others underperform. When every video on a channel gets nearly identical view counts, it indicates systematic view management — either through purchasing or algorithmic manipulation.
A creator whose last 20 videos all have exactly 25,000 ± 500 views has a CV of approximately 0.02. This level of uniformity is statistically impossible with organic audiences and is a hallmark of managed view-purchasing programs.
This heuristic is unique to PulseROI. Neither Modash nor HypeAuditor measures performance uniformity. It is particularly effective at catching Indian SMM panel customers who purchase a fixed batch of views (e.g., 25,000) for every new upload.
Organic subscriber growth follows predictable trajectories. A 3-month-old channel with 500,000 subscribers is almost certainly fake unless verifiable viral content exists. This heuristic catches bulk subscriber purchase events.
Gaining 100,000+ subscribers per month without corresponding viral content is statistically implausible through organic growth alone. CAS flags channels where subscriber count growth cannot be explained by content performance.
Indian mobile-first bot farms can produce this growth signature using real carrier-authenticated devices, making them harder to detect than Western datacenter bots. CAS combined with SVR provides a powerful two-signal verification.
Real creators upload consistently based on their production capacity. Content farms and fake channels often upload 20-50 videos in a single week to build a content library, then go silent for months.
An upload gap standard deviation above 30 days indicates erratic posting — long periods of silence followed by content bursts. This pattern is characteristic of content farms that batch-produce low-quality videos.
This heuristic specifically targets Indian content reupload farms that scrape trending videos from other channels, translate them into regional languages, and bulk-upload them to monetize quickly before being flagged.
A raw engagement rate is meaningless without context. PulseROI evaluates every creator against a benchmark adjusted for their specific Tier, Niche, Language, and Content Format.
We use the median (not mean) engagement rate calculated on a per-video basis. This prevents a single viral video from inflating the channel's overall ER. Each video's ER is computed individually, then the median across 30 videos is taken as the channel's true engagement rate.
Mean ER is easily skewed by one 10M-view viral video. Median gives a more honest picture of day-to-day audience engagement. This is the same methodology used by Modash (November 2025 update) and HypeAuditor.
A creator's engagement rate is evaluated against a benchmark that accounts for four dimensions simultaneously: (1) Creator Tier (Nano/Micro/Mid/Macro/Mega), (2) Content Niche (Tech/Beauty/Finance/Gaming/etc.), (3) Content Language (English/Hindi/Tamil/Telugu/etc.), (4) Video Format (Long-form vs. Shorts). This prevents false comparisons between a Hindi comedy nano-creator and an English tech mega-creator.
No single ER number is universally 'good' or 'bad.' A 2% ER is excellent for a mega-influencer in tech but catastrophic for a nano-creator in beauty. The 4D matrix provides the correct reference frame for every comparison.
After computing the creator's median ER and the adjusted benchmark, we classify the result into five categories: Below Average (< 60% of benchmark), Average (60-100%), Good (100-150%), Excellent (150-200%), and Suspicious (> 200%). The 'Suspicious' category triggers a cross-check against fake engagement heuristics.
An ER that is 200%+ above the expected benchmark is often more concerning than impressive — it frequently indicates WhatsApp engagement pods or coordinated bot activity rather than genuinely exceptional content.
Global tools estimate sponsorship rates using US CPM benchmarks ($15-40 per 1K views). PulseROI uses India-specific Cost-Per-View (CPV) benchmarks calibrated by niche, language, and audience geography.
Every niche in India has a documented Cost-Per-View (CPV) range based on advertiser demand, audience purchasing power, and content format. Finance creators command ₹0.30-0.80 per view. Gaming creators command ₹0.10-0.25 per view. We multiply the creator's average views by the niche-appropriate CPV to calculate the fair sponsorship rate in INR.
After computing the CPV base rate, we adjust for engagement quality. A creator with an ER 1.5× above their niche benchmark gets a 30% price premium (justified by higher audience activity). A creator with ER below 70% of benchmark gets a 40% discount (reflecting lower audience quality). Format (dedicated video vs. integration vs. mention) and language (English premium vs. regional baseline) are applied as sequential multipliers.
Five automated compliance checks run continuously on every monitored video and channel. All checks use the YouTube Data API v3 and cost minimal quota (1 unit per 50 videos for most operations).
YouTube Data API v3 → videos.list with video ID. If the items array returns empty, the video has been deleted or made private.
We batch up to 50 video IDs per API call (1 quota unit per 50 videos). For each monitored video, we compare the current API response against the stored baseline. A missing video triggers a 'video_deleted' alert with severity: critical. We also detect privacy status changes (public → unlisted) by checking the status.privacyStatus field.
YouTube Data API v3 → videos.list with part=snippet returns the full, untruncated description text. We store the description at Day 0 and compare it on every subsequent check.
We parse the stored description for the brand's expected UTM tracking link using case-insensitive string matching (with protocol stripping). If the link was present at Day 0 but absent in the current check, we trigger a 'link_removed' alert with severity: critical. If the link was never present, we trigger a 'link_missing' alert with severity: warning.
Daily snapshots of videos.list statistics.viewCount stored in a time-series table. We compute the velocity curve (daily view gain) and flag statistical anomalies.
Two anomaly types are detected: (1) Cliff-Drop — 90%+ of total views arriving in the first 48 hours followed by near-zero long-tail traffic, indicating a purchased viewburst. (2) View Count Decrease — if today's view count is lower than yesterday's by more than 5%, YouTube has purged bot views from the video. Both patterns generate 'view_anomaly' alerts.
YouTube Data API v3 → playlistItems.list on the creator's uploads playlist. New uploads are fetched and their titles + descriptions are scanned for competitor brand names using AI classification.
We use the Groq AI provider (Llama 3.3, JSON mode) to classify whether a new video's title and description mention any of the brand's declared competitor names. The AI returns a confidence score (0-100). Only detections with confidence above 80% trigger an 'exclusivity_violation' alert. Community Posts cannot be monitored (YouTube API limitation — disclosed to users).
YouTube Data API v3 → playlistItems.list on the creator's uploads playlist (UC → UU prefix swap). Costs only 1 API unit per channel per check.
When a new video is detected on a monitored channel, we fetch its full metadata (title, description, tags) and run keyword matching against the brand's declared brand keywords. Matches are auto-classified as 'brand_video' and automatically registered for 24/7 video monitoring (deletion, link, velocity). Non-matches are logged as 'organic' uploads for trend tracking.
Every analyzed video receives a Brand Safety Score (0-100) computed from three signal layers: ASCI disclosure compliance, GARM content classification, and competitor brand mentions.
We scan the first 200 characters of the video description and the video title for ASCI-mandated disclosure phrases in both English ('Paid Partnership', 'Sponsored', '#ad', '#collab') and Hindi ('पेड पार्टनरशिप', 'प्रायोजित', 'विज्ञापन'). Disclosure must appear before the YouTube 'Show More' fold to be considered compliant.
Compliant (disclosure found in first 200 chars) → 0 penalty. Partial (disclosure found but deep in description) → -10 points. Non-compliant (no disclosure found) → -25 points.
We classify the video's title and description against the Global Alliance for Responsible Media's 11 brand safety categories using AI (Groq, JSON mode). Categories include: Adult Content, Arms, Crime, Death/Injury, Piracy, Hate Speech, Profanity, Drugs/Alcohol, Spam, Terrorism, and Misinformation. Each category is rated at 4 severity tiers: Floor (absolute violation), High Risk, Medium Risk, Low Risk.
Floor/High Risk flags → -20 points each. Medium Risk flags → -8 points each. India-specific sub-categories (caste content, communal content, SEBI financial compliance) are included as additional flags.
We scan the video title, description, and tags for competitor brand names provided by the user during campaign registration. Both exact matches and common misspellings are checked.
Any competitor mention → -15 points. Multiple competitor mentions → -15 per unique competitor.
14-day free trial. 3 monitor slots. All 8 authenticity heuristics active. No credit card required.