How to check a YouTube channel for fake views before you book a sponsorship (with your AI agent)

A sponsorship is priced on views. If some of those views were bought, you are paying for an audience that will never see your ad, let alone buy from you. The view counter can’t tell you. A bought view and a real view add 1 to the same number.

The good news: fake views leave traces. Real audiences comment, like and subscribe in patterns that are hard to fake at scale, and their views pile up on a video in a recognizable shape. This post shows how to check a channel two ways: with your AI agent running the ThoughtLeaders channel authenticity skill, and by hand on YouTube with a short checklist.

What you need

• An AI agent that can run tools: Claude Code, Codex, Gemini CLI, OpenCode or Cursor.
• The ThoughtLeaders CLI connected to it. Setup takes a couple of minutes: see install.
• yt-dlp installed (pip install yt-dlp). The skill uses it to read comments. No cookies or API key needed.
• The channel: a handle, URL, channel ID or name. If you already ran a deal, the sponsorship ID works too.

Step 1: Give your agent the channel

No commands needed. Ask the way you’d ask a colleague:

Vet this YouTube channel for fake views and bot comments before I sponsor it: youtube.com/@channelhandle

If the name matches more than one channel (common for channels with localized copies), the skill stops and shows you the candidates, biggest first, instead of guessing. Pick one and it carries on.

Step 2: What the check actually looks at

Every run does the full check. There are no light modes. The signals fall into three groups.

Engagement compared with similar channels

The skill measures likes and comments per view, then compares them with a peer group of channels in the same niche, not a generic benchmark. A cooking channel and a gaming channel have very different normal comment rates, so “low engagement” only means something next to channels like it. It also sanity-checks audience size across long-form videos and Shorts: a channel whose long-form views dwarf everything else it does deserves a closer look.

View curves and missing videos

Real views arrive in a familiar shape: a fast start, then a slowing climb as the video ages. The skill reads each video’s view history from ThoughtLeaders’ own snapshots (the dataset holds 8.5B of them across the 1.3M channels we track) and looks for curves that don’t behave like organic growth:

• Bursts with no engagement. Views jump, but likes and comments stay flat.
• Cliffs at round numbers. A video that races to just past a round figure and then stops dead, the classic shape of views bought to hit a guarantee.
• Frozen likes. The like count stops moving while views keep climbing.
• Views surging while subscribers stay flat.

It also checks for sponsored or high-view videos that were later deleted or unlisted in a way that hides how past deals performed. Ordinary re-uploads are excluded, so a creator who fixed a typo in a video isn’t flagged.

What the comments say

This is the part the skill never skips, because metrics and view curves can be explained away (“the algorithm picked it up”, “we ran ads”). What the audience actually writes is the most direct evidence there is. The skill reads comments from the channel’s latest long-form videos, its most viewed video and its most recently sponsored one, and classifies each comment as organic, generic template, bot-like, promotional or spam. It looks for:

• Too few comments for the number of views.
• Templated and near-duplicate comments. “Great video!” from fifty accounts.
• Language that doesn’t match the channel.
• Bot-like account handles.

When there are comments to read, the classification runs twice and the two passes are combined, so the share of organic comments it reports stays stable from run to run. That matters because modern comment farms write clean, normal-sounding English. Only reading them closely catches it.

Step 3: Read the output

You get back a short report with:

• A score from 0 to 100, the average of the three group scores.
• A verdict band. The skill’s public bands are 90 and up: Clean; 70 and up: Minor flags; 40 and up: Mixed; under 40: Fraud likely.
• Ranked red flags, most serious first.
• The peer comparison, so you can see how the channel’s engagement stacks up against similar channels.

Two findings can override the average and push the verdict to Fraud likely on their own: a comment section where only a small minority of comments read as organic (or one that is effectively dead), and a pattern of sponsored or high-view videos removed in a way that misrepresents past performance. The skill’s documentation spells out exactly when each one triggers.

One rule for reading it: the score is a reason to check, not a conviction. A low score tells you where to look and what to ask the creator. It doesn’t prove the creator bought views. Channels get hit by bot attacks they didn’t order, and a burst of views can come from a real ad campaign.

Step 4: Act on the verdict

Clean. Authenticity isn’t the question anymore. Whether the channel fits your audience and is worth the price still is, so keep evaluating the proposal on value. When you book, write delivery terms into the deal anyway, because a clean history doesn’t guarantee a clean next video.

Minor flags. Ask for analytics before you sign. Request YouTube Studio screenshots for the flagged videos, covering:

• traffic sources,
• audience geography,
• engaged views,
• returning versus new viewers.

These take a creator a few minutes to pull, and they answer most of what the flags raise.

Mixed. Don’t pay for views you can’t verify. Add a views guarantee: an agreed number of engaged views by a fixed age, such as day 30, counted in the creator’s YouTube Studio rather than the public counter, with a make-good (a bonus mention or a partial refund) if the video falls short. Pair it with the analytics request above. If you want to check the price as well, the quote-check workflow turns the quote into a cost per thousand views you can negotiate with.

Fraud likely. Walk away, or at minimum don’t book until the creator can explain the flags with their own analytics. There are plenty of channels in every niche. You don’t need this one.

Step 5: Audit a deal after it runs

Bought views don’t only happen before a deal. The red flag to watch for after delivery is a sponsored video that stalls short of its guarantee, then jumps over the line just before the deadline. Point the skill at the sponsorship itself (its ID in ThoughtLeaders):

Audit the delivery of sponsorship adlink:12345. Were the views and comments we paid for actually real?

The skill drills down into that sponsorship, so you have something concrete to bring to a make-good conversation.

What a red-flag pattern looks like

We won’t name a channel here, because a pattern is a reason to ask questions, not a verdict on a person. But this is the shape that should make you stop:

• A channel’s recent long-form videos sit at a steady view level, then one sponsored video runs well ahead of the rest.
• Its comments are short, generic praise that would fit any video (“Amazing content”, “Keep it up”), several from accounts with random-looking handles.
• Likes stop moving partway through the video’s life while views keep climbing.
• The extra views carry the video just past a round number and then flatten.

Any one of these has an innocent explanation. All four together, on the video a brand paid for, is the pattern bought-view campaigns leave behind.

The manual checklist: what you can check on YouTube yourself

Most of these need only YouTube; a couple go faster with the free extension, a public stats tracker or the creator’s own Studio data. Open the channel and look at its last 10 to 15 videos. None of these is proof on its own. Treat each as a reason to look closer.

1. View-to-comment ratio. Divide views by comments for each recent video. Compare with two or three similar channels in the same niche, since normal rates vary a lot by category. A channel that gets far fewer comments per view than its peers, or whose ratio swings wildly between videos, deserves the next four checks.
2. Comment quality. Read the comments, not the count. Real comments tend to mention specific moments, argue, ask questions and reply to each other. Generic praise exists on every channel, but a section dominated by short comments that could sit under any video, often posted within minutes of each other, is a flag.
3. View curve shape. Most videos get the bulk of their first-month views early, then slow down. With the ThoughtLeaders Chrome extension installed, the Video Views Over Time panel shows the curve for videos it tracks. Look for steps, late surges with no new reason behind them, and sudden flat lines.
4. Sudden subscriber spikes. Check the channel’s subscriber history in a public stats tracker. Organic channels grow in waves tied to hit videos. A big jump with no video behind it is a flag.
5. Geography mismatch. Compare where the viewers are with who the content is made for and who you’re trying to reach. A channel aimed at US buyers whose views come mostly from unrelated markets needs an explanation. Ask the creator for Studio’s geography report, or compare with the Audience demographics panel in the extension where it has data. Comments in a language the channel doesn’t use are a tell too.

One caveat before you compare numbers. On August 24, 2026, YouTube changed public view counts to count a view from the first frame. Our measurement shows long-form public counts running about 1.2 to 1.5 times the old basis in a video’s first week (Shorts didn’t change). So new long-form videos will look like they have more views per comment than older ones, and public views will run ahead of the engaged views in a creator’s Studio. That gap is YouTube’s counting rule, not fraud. Compare views at the same video age on the same counting basis, keep in mind that older videos keep picking up views under the new rule, and ask for engaged views when you write a guarantee.

FAQ

How can you tell if a YouTube channel has fake views?

Look for views that don’t come with the things real viewers do. Real audiences tend to comment, like and subscribe roughly in line with how many of them there are, and views usually arrive in a curve that slows as a video ages. Warning signs are far fewer comments per view than similar channels, generic or duplicated comments, views that jump without likes moving, and subscriber spikes with no hit video behind them.

Is there a free YouTube fake views checker?

You can run most of the manual checklist above for free. The ThoughtLeaders channel authenticity skill does the heavy lifting: it compares engagement with niche peers, checks view curves and removed videos, classifies the comments, and returns a 0 to 100 score with ranked red flags. It runs inside your AI agent with the ThoughtLeaders CLI.

Does a low authenticity score mean the creator bought views?

No. The score is a reason to check, not a conviction. Channels can be hit by bot comments they didn’t order, and real ad campaigns can produce view bursts. Use the flags to decide what to ask the creator for, usually Studio screenshots of traffic sources and geography.

Can a creator fake YouTube Studio analytics screenshots?

Screenshots can be edited, which is why they work best alongside the checks in this post rather than instead of them. For a large deal, ask for a short screen share of Studio, or write a views guarantee counted in engaged views so the risk sits with the delivery, not your trust.

What should I do if views spike right before a guarantee deadline?

Ask for the video’s traffic sources in Studio for the days of the spike, and run the authenticity skill on the sponsorship to check the curve and comments. If the views came from sources the creator can’t explain, use the make-good terms in your deal.

Try it before your next booking

The fastest way to see whether this is useful is to run it on a channel you’re about to book. Install the CLI and paste the prompt from Step 1 into your agent, or add the Chrome extension and open the channel’s latest videos on YouTube.