AI Companies Place Bulk Orders with Used Bookstores, Twitch Becomes a Learning Resource—The Day When ‘Your Content’ Becomes Worth Zero

Your writings, videos, and accumulated know-how that took you ten years to create are now heading towards a value of zero.

AI companies are on the move seeking training data, and their reach has extended even to used bookstores in the UK. The issue is not about “what will happen to copyright.” It’s about the very structure of content value changing.

—

The Identity of the “Mysterious Bulk Orders” Arriving at Used Bookstores

An unusual phenomenon is occurring in used bookstores across the UK and Ireland. Unknown buyers from the US, Canada, and various parts of Europe are placing large orders—dozens or even hundreds of books at a time. The genres are diverse, making it too unnatural for resale purposes.

It is believed that these orders are part of data collection by AI companies. Reports indicate that Anthropic is purchasing and scanning books worth millions of dollars. The reasoning seems to be that it’s quicker to buy than to get embroiled in copyright disputes.

What we should consider here is the fact that a used book costing a few dollars is “too cheap” for AI companies. Even if they buy 10,000 books at $5 each, that totals $50,000—about 7.5 million yen. Given the value of this data for LLM training, this is an astonishingly low price. Against the backdrop of model development costs in the billions of yen, the data acquisition cost is just a few million yen. There is a discrepancy of over 100 times in value perception between sellers and buyers.

Owners of used bookstores are perplexed. However, the fundamental issue is not just confusion. Your knowledge, your writing, your know-how are being siphoned off at $5 per book and converted into corporate value worth hundreds of billions of yen. There is almost no compensation for you.

—

Twitch Videos Also Become “Learning Materials”

It’s not just books. The same structure is beginning to unfold on Amazon’s Twitch platform.

Twitch has about 100 million monthly active users, with hundreds of thousands of streamers producing content daily. Gaming commentary, chatting, music, cooking—real human behavior across all genres is being accumulated as vast data.

This content is now being considered as potential training material for AI models, leading many streamers to choose to opt out. This is understandable. The videos they have spent thousands of hours creating are being consumed for free by AI, with only the platform and AI companies reaping the profits. The effort put in by creators and the profits gained are completely separated.

Moreover, the troubling aspect is that opting out may not be effective. Once content is published online, it can be easily scraped. With just a line in the terms of service stating “we will use this for AI training,” the current legal framework allows companies to bypass gray areas.

—

The Problem of “Quality Compression” Shown by AI-Generated Novels

So, what about the quality of the content created using data absorbed by AI?

Recent studies have shown that novels generated by LLMs (large language models) exhibit clear characteristics. The structure of sentences becomes uniform, the variation in readability narrows, and the use of punctuation becomes standardized. In short, “they are readable, but they all taste the same.”

Even when mimicking the style of 19th-century realism, while the superficial style can be reproduced, the rhythm of the writing, the ebb and flow, and the “tension” that surprises readers disappear. Writing that humans read and feel “it’s good, but something feels off” is mass-produced.

This issue is not limited to novels. The same phenomenon occurs with blog posts, manuals, proposals—any type of text content. The cost of AI-generated content approaches near zero. What used to cost 30,000 yen per article can now be reduced to a few dozen yen through API usage. When costs drop to one-thousandth, what happens to the “market price” of that content? The answer is clear.

—

In a World Where the Price of Content Becomes Zero, What Will Still Hold Value?

After reading this, you might wonder, “Does this mean the end for creators?” No. What is ending is the era where ‘creating content itself’ had a price.

Let’s clarify. There are things whose value decreases due to AI and things whose value increases.

Things whose value decreases:

• Content that is merely “summarized information”
• Formulaic writing that anyone could produce
• Stock-type content that competes on quantity
• Productions made just for the sake of making something

Things whose value increases:

• Primary information that can only be obtained in specific contexts
• Personal trust that holds meaning because “this person said it”
• Latest experiential data that AI has not learned
• Context within small communities

Here lies an opportunity for small and medium-sized enterprises.

—

Three Ways for SMEs to Break Free from Being “Sucked Dry”

There is no need to compete on the same stage as large corporations. In fact, there are things that only small and medium-sized enterprises can do.

1. Recognize Your Company’s Primary Information as a “Valuable Asset”

The data present in your company—construction records, customer interaction logs, industry-specific know-how—are pieces of information that AI cannot pick up from the internet. This means they have rarity.

For example, if a local construction company structures 20 years’ worth of construction trouble cases into a database, this holds extremely high value as both AI training data and educational material within the industry. Data that cannot be purchased from outside can be priced.

2. Transform Content from “Consumable” to “Relationship-Building”

AI can mass-produce content that is “read and done.” Conversely, content that makes readers think, “I want to consult this company” or “I want to meet this person” cannot be created by AI.

Specifically, it involves sharing your company’s stories of failures and the processes of trial and error. Instead of saying, “This is how we succeeded,” share, “This is how we failed and how we fixed it.” This holds a completely different value from general knowledge found through searches.

3. Shift to Being on the “Utilizing AI Side”

Instead of lamenting as the “sucked dry” side, it’s faster to turn to being the ones who use AI.

Creating internal manuals, drafting initial responses to customer inquiries, drafting estimates—if tasks that previously took two hours can now be completed in 15 minutes, the time saved can be focused on “things that AI cannot do.” Go out into the field, talk to customers, try new construction methods. Instead of losing jobs to AI, delegate mundane tasks to AI and enhance the value of human work.

—

So, What Should We Do?

The stories of used bookstores and Twitch are fundamentally the same. “Don’t let others decide the price of what you possess.”

AI companies will buy your content for $5 a piece. Platforms will use your videos as free learning materials. If you remain silent, the price will converge to zero.

However, insights that can only be gained on-site, trust relationships with customers, and context rooted in the community—these are things that AI cannot generate. They cannot be scraped either. Rare things hold value.

What needs to be done is threefold: inventory the value of your company’s primary information, transform content from “consumption” to “relationship building,” and shift to being on the utilizing AI side.

The era where “the price of content becomes zero” is simultaneously an era where “the price of things that do not become zero skyrockets.” What are the “things that do not become zero” that your company possesses? Identifying this is the most important task right now.