Can This Go Live? #2

What is this blogpost about?

This is another addition to the blogpost series “Can This Go Live”. The first blogpost outlining the series is here.

Published: 2025 July 4
Disclaimer: The following is an educational analysis based on the author's interpretation of publicly available data using the framework outlined in this post. It constitutes the author's opinion and is not a definitive assertion of fact.

What stumbled upon?

About the video

  • Platform: YouTube
  • Title: BLACK PANTHER EYES OF WAKANDA TRAILER 2025: Welcome Back Iron Fist
  • Creator: Emergency Awesome

About my interest in the video: The fictional world of the Marvel universe has captivated me since my childhood. I would watch the cartoons whenever they aired on TV, I would borrow the VHS tapes I could find on a shelf, and I would read the comics I could buy thanks to the generosity of my parents. To make sure that I noticed and understood all of those “easter eggs”, I started watching Emergency Awesome. Since first watching the channel’s videos ~10 years ago, I eventually hit the channel’s subscribe button, and I have clicked into almost all videos ever since.

About how I stumbled upon the examples

After writing “Can This Go Live #1” and while watching the video, I thought to myself, “Could there be more unexpected behaviors in the comments section?” Following this thought, I proceeded to do my routine scrolling. While scrolling, I sorted the comments by the “Newest first” option. At the bottom of the section, I came across a group of 6 accounts (G1).

Behavioral Properties

G1 piqued my interest for further exploration because of the following behavior-dimension properties:

  • Coordinated Timing: 6 of 6 accounts posted a comment within a 1 to 2 minute time period.
  • Content Structure: 6 of 6 accounts posted a comment which (i) had the sentiment of gratefulness for the creator’s work, (ii) had a similar content structure of 2 sentences followed by 3 emojis, (iii) had gibberish emoji sequences.

Detailed Account Analysis

Moving onto account-dimension properties:

  1. 6 of 6 accounts had an account profile picture of an attractive, possibly eye-catching, female. In addition, 2 of 6 accounts had what appears to be the same profile picture.
  2. 6 of 6 accounts had an account profile page which included eye-catching text to describe links to pairs of on-platform accounts (A1, A2) and (A3, A4). Refer to these as G2.
  3. G2 accounts all have account profile pictures of a faceless female with an emphasis on provocativeness.
  4. G2 accounts all have account profile names which could be categorized as part of the sexual content category.
  5. G2 accounts all have outbound links to off-platform digital products/services hosted in the “beacons.ai” domain.
  6. All accounts reviewed above were created within 2 days of the video’s publication date.
Figure 1
Figure 1: Coordinated comment timing and structure
Figure 2
Figure 2: Profile picture analysis
Figure 3
Figure 3: Cross-profile link consistency

Summary of Problematic Activity

Based on the exploration, there is reason to believe that a group of coordinated, inauthentic accounts (G1) acted to increase traffic to off-platform products/services (G2) by using outbound links. G1 accounts posted comments soon after publication, with positive sentiment to initiate engagement, using eye-catching pictures and text to lead users to G2 accounts, which then provide the final off-platform links.

Conclusion: Coordinated, inauthentic accounts using attractive profile pictures and suggestive text to drive traffic to suspicious Beacons.ai landing pages.

Thoughts & Next Steps

Community Guidelines

These examples may violate YouTube's “Spam & deceptive practices” and “Sensitive content” policies. Specifically, the “Repetitive comments” guideline and “External links policy” are relevant, as well as “Nudity & Sexual Content Policy”.

Detection Workflow Thoughts

  1. What is the chosen approach for detecting similarity between comment posts, pictures, descriptions and usernames?
  2. What are the alternative approaches and tradeoffs?
  3. What is the scale of these problematic activities?
  4. Are there comparable behaviors on other platforms?
  5. What are the impacts on content consumers, producers, and advertisers?

Actions to Take

  • For Analysts: Consider how these activities may take form on the platforms you are responsible for.
  • For Creators: Use "Blocked words" functionality and request platform enhancements.
  • For Users: File reports to help platforms realize the relevance of removing such behaviors.
Disclosures:
  1. I worked in the Trust & Safety for 7 years. In 0 of those years, I worked on the YouTube UGVC platform.
  2. I interview for open roles occasionally with some of the companies written about.
  3. I own equity in some of the companies written about.

Footnotes:
  1. I do mean “clicked” and not “watched”. I click into the videos to read the comments section.
  2. Account names not redacted due to the significant nature of the sexualized keywords used. The risk of harassment/doxxing is considered very low.