Title

LLMs All Say the Same Thing—The Only Way for SMEs to Differentiate in the Era of ‘AI Homogenization’

Body

If Everyone Uses the Same AI, Everyone Gets the Same Answer. You Can’t Win That Way.

You ask ChatGPT to come up with a catchphrase. You ask Claude. You ask Gemini. The phrases that come out are astonishingly similar.

This is not just a coincidence. It’s a structural problem.

An experiment reported by MIT Technology Review illustrates this clearly. When asked to “pick a random number from 1 to 10,” nearly all major chatbots respond with “7.” It’s well-known that humans also tend to say 7 when asked the same question, but AI learns this from its training data and reproduces it even more strongly. It’s not even random.

In a NeurIPS award-winning paper, when instructed to “write a metaphor about time,” 25 different LLMs mostly responded with “Time is a river.” Twenty-five models, the same metaphor. This is the current reality of AI.

The issue isn’t just about literary sensibility. If your company’s marketing copy, proposals, and customer response emails are all generated using the same AI, the outputs will be the same. If you fight with the same weapons, there will be no differentiation. If there’s no differentiation, you can only compete on price. For SMEs, that spells disaster.

The Pitfall of Thinking “AI Will Summarize for Me”

The homogenization of AI is not just a problem of similar outputs. What’s more frightening is that “important information gets lost.”

There’s a notorious case involving AI-generated review summaries on TripAdvisor. Despite multiple reports from users about sexual violence and food poisoning at a particular hotel, the AI-generated summary rated it as “spotless.”

Why does this happen? LLMs are influenced by the majority opinion. If 95 out of 100 reviews say “clean and comfortable,” the remaining five serious allegations are statistically dismissed as noise. They get averaged out.

Let’s translate this to the context of SMEs. You have AI summarize customer survey open-ended responses. Out of 100 responses, only three say, “The service was terrible” or “I will never use this again.” What does the AI summary say? “Overall positive feedback. Many customers are satisfied with the quality of service”—and that’s how it goes.

Those three responses are potential cancellations for next month. Whether you can catch those three could mean hundreds of thousands of yen in annual sales. AI is good at averaging, but what separates the life and death of SMEs is not the average, but the outliers.

What Happens After Costs Go Down

Let’s think structurally for a moment.

With the advent of LLMs, the “cost of writing decent content” has dramatically decreased. A blog post that used to cost 30,000 to 50,000 yen to outsource can now be created with ChatGPT Plus for just 2,000 yen a month, allowing for unlimited production. This is effectively a reduction of over 95% in costs.

This is a wonderful thing. But what happens when costs go down? Everyone does it. When everyone does it, the value of “decent content” approaches zero.

SEO articles are a clear example. After 2024, AI-generated SEO articles exploded in number. According to a study by Originality.ai, about 57% of the content that ranks well shows traces of AI generation. Similar structures, similar headlines, similar conclusions. Google updated its algorithms to respond to this homogenization, and in the core update of March 2024, it excluded a large number of low-quality AI articles from its index.

As the cost of “decent” decreases, the relative value of “genuine” increases. This is the current structure.

The Reverse Structure Where SMEs Can Win

Now, let’s get to the main point. “So what should we do?”

Large companies are more prone to fall into the trap of homogenization. Why? Because there are many people involved in decision-making, and it’s easier to pass AI outputs as “safe and acceptable.” Brand guidelines, legal checks, approvals from multiple departments—going through that process, sharp expressions get trimmed, and ultimately, the “decent” text generated by AI is what gets published.

SMEs are different. If the CEO thinks, “This is interesting,” it can be published the next day. They can articulate the feel of the field directly in words. What AI can never generate is “specific experiences that can only be told by this company.”

For example, a local sheet metal processing company started a blog series titled “The Most Difficult Processing We’ve Done in Our Factory.” They write specifically about cases where craftsmen faced real challenges, complete with photos. AI can’t write that. Why? Because that experience doesn’t exist anywhere on the internet. As a result, inquiries through this blog increased by 2 to 3 per month. Considering the unit price of sheet metal processing, that’s an impact of several hundred thousand to a million yen per month. The cost is almost zero. The CEO writes it on their smartphone during lunch breaks.

As AI homogenizes, the rarity of “primary information unique to that company” increases. This is the reverse structure.

Practice: Three Rules to Turn AI into a “Differentiation Tool”

I don’t want to end with abstract discussions, so here are three specific rules.

Rule 1: Don’t Use AI Outputs as They Are. Always Mix in “Your Company’s Facts”

It’s fine to have AI draft a newsletter. However, include one episode that only your company has, like “Last week, I heard from Mr. △△ in 〇〇 City.” Just that will make it a different piece from other companies’ AI-generated newsletters. The cost increase is almost zero. It takes just five minutes.

Rule 2: Don’t Take AI Summaries at Face Value. Look at the “Outliers” with Your Own Eyes

Customer surveys, reviews, inquiry logs—having AI summarize these is efficient. However, you shouldn’t be satisfied with just reading the summary. Once a month, set aside 30 minutes to read the raw data yourself. The “outliers” that AI discards hold the next business opportunities and risks.

Rule 3: Invest the Time Saved by AI into “Things Only Humans Can Do”

Let’s say you automate administrative tasks with AI and save 20 hours a month. What will you do with that 20 hours? Automate another task with AI? No. Go meet customers. Go see the field. Gather “primary information” that AI can’t generate. Reinvest the resources saved from cost reduction into the source of differentiation. This is the essence of AI utilization for SMEs.

It’s Not “Don’t Use AI”; It’s “Don’t Become the Same as AI”

Let me clarify to avoid misunderstandings. This is not about not using AI. Rather, SMEs should thoroughly leverage AI. By bringing the costs of routine tasks, data organization, draft creation, and meeting minutes summarization close to zero, they can bridge the gap in resources with large companies.

However, you must not make AI outputs your company’s “face.” In an era where everyone uses the same AI, AI outputs are the “starting line,” not the “goal.”

I want to pose a question. The text on your company’s website, proposals, and customer response emails—if you hide your company name, can you distinguish them from your competitors’?

If you can’t tell them apart, then you haven’t “had AI do the work”; you’ve merely “become the same as AI.”

The weapons of SMEs are speed, field sensitivity, and the decision-making ability of the management. AI is a whetstone to sharpen those weapons, not the weapons themselves. If you forget to wield your sword relying on the whetstone, you won’t win even the battles you could have.

Here’s something you can do starting tomorrow. Open one piece of text written by AI and ask, “Does it contain even one piece of information that only we could write?” If it doesn’t, add it. Just that will help your company step out of the herd of “groupthink AI.”