YouTube fake views checker
Paste a link to a YouTube channel or video. In a few seconds you'll see whether its views, likes and comments look like a real audience. Useful before you pay a creator for an ad.
- Up to 50 recent videos, up to 300 comments and their authors
- Each channel is compared with its own videos and its category
- Every finding comes with the number and the usual range
What we check
One number rarely proves anything, so we look at four groups of signals. A high risk needs at least one strong signal. The top verdict needs strong signals from different groups.
Channel
Lifetime views per subscriber, how unevenly the videos perform, sudden jumps in views.
Videos
Likes and comments per view, and how much they vary between videos. Real audiences are uneven. Bought engagement tends to be flat.
Comments
Repeated templates, generic praise, comments in another language (Russian-language channels), bursts within 10 minutes. A language model also reads a sample and counts the comments about the video itself.
Commenters
How new the accounts are, whether many were created in the same month, and how many come from countries known for click farms (Russian-language channels).
How to read the score
| 0-24 | No signs of fake engagement |
| 25-49 | Engagement raises questions |
| 50-74 | High risk of fake engagement |
| 75-100 | Very likely fake engagement |
Small deviations on their own never push a channel to high risk. We don't score channels with fewer than 10,000 subscribers or too few videos, because there isn't enough data for a fair result.
What the result means
- We only see public data. Traffic sources and watch time are visible only to the channel owner in YouTube Studio.
- Cheap Google Ads views can look like bought views from the outside: many views, few reactions. For an advertiser the outcome is similar.
- Someone else can bot a channel without the owner's knowledge. If in doubt, ask the creator for screenshots from YouTube Studio.
Methodology
Each check computes up to 32 signals from public YouTube data. Every signal has a usual range. If a channel falls outside it, the signal adds half or all of its weight to the score. Signals marked “strong” can raise the result above “Engagement raises questions”. The numbers below are the thresholds we use now.
Channel
Lifetime views per subscriberstrong
Suspicious below 60, very suspicious below 30. Both limits are halved for channels under a year old. Bought subscribers don't watch videos.
Spread of views across videosstrong
Interquartile range of views divided by their median. We use long videos, and Shorts on channels made almost only of Shorts. Suspicious below 0.5, very suspicious below 0.3. Real audiences watch some videos far more than others.
Median views vs subscribers
Suspicious above 60% for channels with 100,000+ subscribers, or below 3%. Shorts channels have no upper limit and half the lower ones: Shorts reach does not depend on subscribers.
Jump in views
Median views of the newer half of videos compared with the older half, in the same format (long videos or Shorts). Suspicious above ×5, very suspicious above ×10.
Other channel checks
Average views per video almost equal to the median on small channels; long videos far ahead of Shorts.
Videos
Likes per viewstrong
A median below 1% is suspicious, below 0.5% very suspicious. Half these limits for music. If comments are off, this signal can only add a warning.
Comments per 1,000 viewsstrong
A median below 0.5 is suspicious, below 0.2 very suspicious. Adjusted by category: music ×0.5, gaming ×0.8, news ×1.2, Shorts ×0.4.
Flat engagement
Like and comment rates barely change from video to video (coefficient of variation below 0.35).
Round numbers and mismatches
Views stuck just above a round number, normal likes with almost no comments or the reverse, views removed between two checks.
A single video
For a video link we compare its views and engagement with the channel medians.
Comments
Repeated commentsstrong
Share of comments that are identical or nearly identical. Suspicious above 12%, very suspicious above 20%.
Comments about the videostrong
A language model reads up to 50 comments and counts the ones about the video itself. Suspicious below 35%, very suspicious below 25%. We run it only when the other signals leave doubt.
Templates from other channelsstrong
Longer comments whose exact text also appears on 3 to 100 other checked channels. Suspicious above 5%. Turns on once 200 channels are in our base.
Praise, timing, language, links
Generic praise, bursts within 10 minutes, comments in another language (Russian-language channels), links and “DM me”, almost no likes on the top comments.
Latin letters inside Russian wordsstrong
Share of comments where Russian words contain look-alike Latin letters. Bots do this to slip past filters and duplicate search. Suspicious above 5%, very suspicious above 10%. From 20% of such comments the score is at least 50.
Commenters
Accounts under 90 days oldstrong
Suspicious above 20%, very suspicious above 35%.
Accounts created in the same monthstrong
The largest share of older accounts created within one 30-day window. Suspicious above 20%, very suspicious above 30%.
Authors from flagged channelsstrong
People we've already seen under channels our team marked as fake by hand.
Profiles and location
Empty accounts, default avatars, the same people under every video, commenters from countries known for click farms (Russian-language channels).
How the score works
- Each signal adds 0, half or all of its weight. The total runs from 0 to 100. Signals we couldn't compute are skipped: they lower the confidence, not the score.
- Without at least one strong signal at full weight, the score stays at 49 or below.
- Some combinations set a minimum of 50: the same comment under several videos (5+ copies, 3%+ of the sample, 3+ authors), 5+ templates seen on other channels, new accounts from click-farm countries, fewer than 10 lifetime views per subscriber on an established channel, or few views per subscriber together with very even views across videos, or 20%+ of comments with Latin letters inside Russian words.
- A minimum of 75 (“Very likely fake engagement”) applies only when different groups agree. For example, low views per subscriber backed by comment or commenter signals.
When there's no score
- Fewer than 10,000 subscribers. If the count is hidden, we use the median views of long videos, or of Shorts for Shorts channels, and need at least 2,000.
- Fewer than 8 videos older than 14 days. Newer videos and live streams don't count toward the ratios.
- We could compute less than half of the signal weight.
- Confidence is high when both comment and commenter signals are computed and cover at least 80% of the weight.
Data and privacy
- We use public YouTube data only: the channel, up to 50 recent uploads (a few more pages if they hold few long videos), the top 100 comments under three videos and up to 150 commenter profiles. On Shorts channels we read the most commented Shorts.
- YouTube returns comments sorted by relevance, so the sample leans toward popular comments, which are usually real. Bots further down the list count for less than they should.
- Commenters are stored only as salted hashes, and comment texts aren't saved in reports. Reports are kept for 90 days. A repeat check within 7 days comes from cache.
- The language model runs on our own servers. Comments aren't sent to third parties.
Methodology version yt-check-0.7, 4 November 2026. We review the thresholds on hand-labeled channels. The score estimates risk. It isn't proof.
Need a full creator check?
Yoloco shows audience quality, real engagement, ad history and similar channels on YouTube, Telegram, VK and Instagram.
Questions about the checker
How do I check a YouTube channel for fake views?
Paste a link to the channel or to any of its videos into the form above. We collect public stats for recent videos, comments and commenter profiles, then show a risk score with an explanation for each signal.
What gives fake views away?
Few lifetime views for a large subscriber count, almost identical views on every video, very low likes and comments per view, template comments from new accounts. One signal on its own proves little. It matters when several point the same way.
Is it free?
Yes, and you don't need to sign up. You can also get the full breakdown with every signal by email.
How accurate is it?
The checker is built to catch crude inflation: networks of channels with bought views and template comments. Careful boosting on top of a real audience isn't visible in public data, and we don't claim to catch it. We tune the thresholds on channels we've labeled by hand. Treat the score as a risk estimate, not an accusation.
Why didn't a channel get a score?
We don't score channels with fewer than 10,000 subscribers, fewer than 8 videos older than two weeks, or with comments turned off. We still show what we could measure.
I own the channel and disagree with the result
Use the “Dispute” link under the result. If you share stats from YouTube Studio, we'll review it by hand.
Find creators with a real audience
Search millions of channels with checked engagement and audience quality. Free to start.
Prefer a walkthrough? .