2 min read

Deepfakes, Short Videos, and Platform Algorithms

Hashtag totals and a recommendation feed do not reveal why each video was shown. Platform labels identify some disclosed synthetic media, while independent checks of the original clip remain essential.

Miriam Bernstein

Miriam Bernstein is the pen name of Hasbara’s Resources desk. Articles on this desk are written and published automatically by artificial intelligence, with no person reviewing them before publication.

Hashtag totals and a personalized feed cannot, by themselves, show why a platform recommended a particular video or whether it systematically favors a viewpoint. Synthetic-media labels can identify some realistic altered or generated content when creators disclose it. Check the video’s source and context independently, whether or not it carries a label.

Read platform figures for what they measure

TikTok’s November 2023 statement said that, in the United States, #StandWithIsrael had 68% more views per video than #FreePalestine at that time, despite fewer videos tagged with it. TikTok also said hashtags are added by creators and that the newer hashtag’s videos were more recently posted. These are figures reported by the platform for that snapshot; they do not establish what an individual viewer was recommended or why. The company’s explanation of its hashtag comparison describes the figures and its interpretation.

To assess a claim about recommendations, look for a study that states when it collected data, which accounts or searches it used, the sample size and what it counted. A single person’s feed reflects that account’s activity and settings; hashtag totals count posts or views, not the causal effect of a recommendation system. Treat broad claims about systematic platform bias as requiring evidence designed to measure that question.

Understand what a synthetic-media label covers

YouTube’s March 2024 creator guidance requires disclosure for realistic altered or synthetic material that viewers could mistake for a real person, place or event. It exempts clearly unrealistic or animated material, special effects, production assistance and other cases described in the policy. The YouTube disclosure announcement gives examples, including a digitally replaced face and altered footage of a real place.

A label is useful evidence about a creator’s disclosure, but it does not authenticate every repost or prove that an unlabeled clip is genuine. Content Credentials can record signed provenance and modifications when present; the C2PA FAQ explains how the credential is linked to the file. An image or video without credentials still needs ordinary source, date and location checks.

Find the earliest version of a suspicious clip, inspect multiple frames, and compare it with independent reporting or original footage. Do not call a video AI-generated simply because it looks unusual. See how to trace recycled imagery and the broader verification routine.

Sources

  1. 1.https://newsroom.tiktok.com/the-truth-about-tiktok-hashtags-and-content-during-the-israel-hamas-war/?lang=en
  2. 2.https://blog.google/intl/en-in/products/platforms/how-were-helping-creators-disclose-altered-or-synthetic-content/
  3. 3.https://c2pa.org/faqs/

Accessibility

These settings are saved on this device.

Preview

Read clearly, your way

Adjust this page while you read. Explore resources

Text
Text sizeApplies across the site.
Colour and contrast
Theme
Motion
AnimationsMatch device follows your reduced motion setting.