In an era of mass video posting and automated content distribution, many companies rely on AI generation for scaling. However, behind the scenes of this industry lie risks that can nullify all efforts. While services for automatic short video publishing promise seamless integration and time savings, major AI developers are buying tons of old printed books. Why? Because books printed before the era of “junk” content possess a quality that cannot be imitated.
Why Are AI Giants Buying Old Books?
- Data Quality: Books published before 2022 contain unspoiled data, free from AI-generated “noise.”
- Avoiding “Junk”: Companies creating AI pay real money to avoid using their own “junk” content for training.
- Irreproducibility: Unlike AI responses, which can change with repeated queries, the quality of old books is stable.
“The best data for training AI is on the shelf,” says ISBNdb, a broker supplying books to AI labs.
Watermarking AI Content: Invisible Control
The industry is on the verge of a new era of control over AI content. On May 19 (at I/O 2026), Google announced that its invisible watermarking system SynthID has already marked over 100 billion AI images and videos, as well as about 60,000 years of audio. Verification of these marks is already being implemented in Google Search and Chrome.
Key Players and Their Commitments:
- Google: SynthID for images, videos, and audio.
- OpenAI: Committed to embedding SynthID in all images generated via ChatGPT, Codex, and API.
- Anthropic: Starting August 2, 2026, Claude models will embed watermarks in generated text at the model level worldwide.
This means that any automatic short video publishing or text will carry an invisible trace of its origin. Even if you use a single dashboard for TikTok, YouTube, Instagram for scheduling auto-posting, your content may be marked.
Limitations of Watermarks and Circumvention Methods
Despite their apparent universality, watermarking systems have their limitations:
- Rewriting and Translation: Deep rewriting or translation of text significantly reduces the accuracy of watermark detection.
Tools for removing watermarks have already appeared, for example, on GitHub, which use text rewriting through another model. However, this does not guarantee complete removal of the mark, and posting to 10 accounts with such content still carries risks.
Why is AI content «junk»?
AI content, created in large volumes, often does not fall into the main body of training data used to create the parametric memory of models. This means that it does not form long-term value and does not contribute to brand recognition.
A geoSurge study showed that models are more likely to look for what they already know. Brands in the top 10 of the model’s memory were mentioned in search queries 3.2 times more often (55.7% vs. 17.4%). This suggests that even publishing on 5 platforms will not help if the content does not get into the AI’s «memory.»
Google Research confirms: frontal models encode 95-98% of facts, but cannot recall a quarter or a third of them directly. This means that even if your AI content is encoded, it may not be accessible without special prompts.
Conclusion: Bet on quality, not quantity
Betting on mass video posting generated by AI is a risk that can lead to a loss of content value. Companies producing AI are actively fighting «junk» content by buying quality data and implementing detection systems.
If you use a mass posting service, remember: real value is not in volume, but in quality and originality. A long-term strategy should be aimed at creating content that will form stable parametric memory in AI models, rather than trying to deceive detection systems. The cost of 500 publications may be zero if your content is labeled as «junk.»
Stay up-to-date with the latest developments in AI and SEO to ensure your monthly mass posting package brings real returns. Learn how to integrate AI and SEO for maximum effectiveness.
Frequently Asked Questions
What is SynthID and how does it affect mass video posting?
SynthID is an invisible watermarking system from Google that tags AI-generated images, videos, and audio. It affects mass video posting as it allows for the identification of AI-created content, which can impact its ranking and distribution.
Can watermarks be removed from AI-generated text?
There are methods for removing watermarks, such as deep rewriting of text or translating it through another model. However, the effectiveness of these methods is not guaranteed, and AI development companies are constantly improving their detection systems.
Why are AI companies buying old printed books?
AI companies are buying old printed books to train their models because these books contain high-quality, unspoiled data created before the advent of mass AI-generated content. This helps improve the quality and reliability of AI systems.
How to choose a mass posting service considering the risks of AI content?
When choosing a mass posting service, prioritize those that focus on content quality, not just volume. Look for solutions that help create unique, valuable content capable of forming lasting memory in AI models, rather than just generating «junk.»
