To conclude, rather than trying to make money by building something with AI, it's more like you should just invest that money in stocks instead 😁

Recently, I've been watching the term GEO, or Generative Engine Optimization, gradually surface in the marketing industry. In simpler terms, it's about optimizing so that when generative AI like ChatGPT, Perplexity, and Gemini—rather than Google search—provide answers, our brand or products get mentioned.

I've actually reviewed ideas in a similar direction before. The beginning was a bit odd. At first, it was an idea close to an "AI agent honeypot." If AI agents will roam the web and find information in the future, I thought it might be possible to preemptively plant commercial or advertising information that those agents could read.

However, upon review, I concluded that simply luring or deceiving AI would be hard to sustain. Eventually, the direction shifted toward GEO. However, it wasn't in the sense of "tricks to make AI recommend our products" as the marketing industry talks about it now.

The conclusion I reached was closer to this.

Structuring and building up product information, price information, comparison information, review information, and market price information that AI must refer to.

In other words, I saw GEO not as a trick, but as something closer to building a product knowledge base that AI can read.

But Why Did I Abandon This Direction?

The reason is simple.

Because I didn't believe it was yet a market that ordinary people could jump into.

There's one thing I need to be clear about here. Throughout 2025, AI shopping has moved much faster than when I first reviewed it.

• OpenAI added shopping recommendations with images, reviews, and purchase links to ChatGPT Search in April 2025, and announced that it uses structured metadata such as prices, product descriptions, and reviews, but does not accept advertising or commissions. (Reuters)

• In September 2025, Instant Checkout was introduced for Etsy and Shopify merchants, and by October with Walmart and November with Target, the flow of direct payment within ChatGPT was added. (Retail Dive)

• On November 24, 2025, the GPT-5 mini-based "shopping research" feature was rolled out to all free, Go, Plus, and Pro users. OpenAI announced that the accuracy of multi-condition product queries increased from 37% to 52% compared to regular ChatGPT Search. (OpenAI)

• Perplexity is operating an Instant Buy feature for US users. According to Perplexity's official Help Center, users can search and purchase products within Perplexity, and sellers are responsible for order fulfillment and after-sales service. (Perplexity Help Center) This feature was combined with PayPal's official partnership on November 25, 2025, bringing together payment, identity verification, and buyer protection infrastructure, with Abercrombie & Fitch, Ashley Furniture, Adorama, NewEgg, and others joining as initial merchants. (PayPal Newsroom)

• According to Adobe Analytics, during the 2025 holiday season, AI-sourced traffic to US retail sites increased 693% year-over-year, and shoppers who came via AI had 33% lower bounce rates and 31% higher conversion rates. (Shopify)

• Google also enabled existing Search, Shopping, and Performance Max campaigns' text and shopping ads to be displayed within AI Overviews. (Google Ads Help)

In other words, it's true that it has begun to become a channel where money flows, beyond just having the functionality. However, this trend and "whether ordinary people can earn money by making GEO a marketing channel" are different stories. The latter remains highly questionable.

AI Shopping Is Still Heavy, and Divided by Category

I saw UX as the biggest variable, and this judgment hasn't changed much even after a year.

Shopping is fundamentally a very fast action. Users enter keywords in the search box, product cards appear, they see the price, look at reviews, check shipping costs, change filters again, and change the sorting. If they're not satisfied, they go back immediately, open other products in new tabs, and check coupon or shipping conditions.

The core of online shopping is not "the right answer" but fast exploration.

When buying things, "the most logical recommendation" is often less important than "quickly skimming and getting a feel for it." This is especially true for categories such as daily necessities, food, clothing, and small household items.

OpenAI itself acknowledges this point. Shopping research is "suitable for in-depth comparisons that take time," while simple price checks or feature checks are faster with a regular ChatGPT response, as stated explicitly (OpenAI).

Therefore, the scope of AI shopping eventually narrows down to this area.

Conversely, in markets where taste, shipping, reviews, coupons, and fit are important, such as daily necessities, food, repeat purchases of low-priced items, and clothing, conversational AI may still find it difficult to replace shopping malls. It's likely that the increase in Adobe Analytics traffic will also be concentrated in categories such as home appliances, home goods, and beauty, where comparison is crucial.

The Point oo.ai Tried to Break Through

In this context, I found oo.ai to be a rather interesting case.

oo.ai seemed less like "conversational AI search" and more like an attempt to reduce the burden of AI search. The core idea I estimated at the time was this:

Instead of making users wait for LLM responses every time, combine AI summaries and ads on top of pre-indexed and cached search results.

In other words, I believed that oo.ai's BM model could be closer to a structure where AI search indexes are embedded with ads or commercial information in the long run, rather than simply a subscription fee. Just as existing search engines combined search indexes and advertising, AI search could ultimately create commercial exposure space between "questions" and "answers."

This direction also aligns with GEO.

However, this model is not easy, both technically and commercially.

AI search is more expensive than regular search, slower in response time, and raises trust issues when ads are attached. Users may lose trust the moment they perceive AI answers as "billboards." Conversely, it's difficult to cover the costs of a large-scale search infrastructure without advertising revenue.

But oo.ai Failed for Another Reason

OpenResearch, which operates oo.ai, is a company founded in July 2024 by Kim Il-doo, a former employee of Kakao Brain. The company attracted attention after securing 10 billion won in seed funding from LB Investment, Mirae Asset Venture Investment, and Bon Angels Venture Partners within two months of its establishment. oo.ai, launched in March 2025, surpassed 2 million MAU (monthly active users) in just nine weeks. (JoongAng Ilbo)

However, the service was temporarily suspended at the end of November 2025. In a statement to Digital Daily, CEO Kim explained that the decision was made for cost efficiency and pivoting, citing weekly expenses of 200-300 million won. He also denied allegations of gambling as "groundless." (Digital Daily)

About ten days later, CEO Kim posted an apology on his social media account, changing his stance.

• He admitted to gambling. He said he first encountered a casino during a Las Vegas trip in April 2024 and lost his judgment due to depression and panic disorder.

• According to him, the gambling funds were not from company investments but from personal funds raised through the sale of assets.

• He acknowledged a procedural violation (lack of shareholder consent) in the disbursement of advance payments.

• He is currently negotiating with investors on matters such as liquidated damages and put option exercises. (AI Times, Nate/Digital Daily)

In essence, while CEO Kim denies "direct embezzlement of company funds," the confirmed facts are his gambling, procedural violations, and initial false explanation. This is the extent of what can be safely stated at this point.

The future of the service itself is clearer. The search function was limited until January 26, 2026, and even after that, it operated inconsistently. Since around April 2026, the oo.ai domain has been redirected to Afternic (GoDaddy), effectively marking the end of the service. The domain itself is listed for sale at $280,000. (Namu Wiki)

The important takeaway from this case is that the idea behind oo.ai was not entirely flawed.

Rather, I believe that the direction oo.ai aimed for - addressing issues such as the speed of AI search, caching and indexing problems, and the potential for advertising models within that framework - remains a significant topic.

Ultimately, oo.ai collapsed due to CEO risk and financial controversy before its technology or business model could be sufficiently validated in the market.

This is unfortunate because oo.ai was one of the few domestic cases that could have verified the actual market potential of AI search and GEO (Generalized Entity Ontology).

GEO may be necessary, but it's not currently a profitable market for the general public

It's not accurate to say that GEO itself is wrong. In fact, its direction may be correct.

Google Merchant Center emphasizes the importance of providing accurate product data. Google states that it uses product data from Merchant Center to match products with relevant search terms and that properly formatted and accurate product data is crucial for both ads and free listing visibility. (Google Merchant Center Help)

ChatGPT's shopping research is similar. OpenAI states that if retailers are added to the allowlist, their product catalogs can be exposed to ChatGPT's 800 million users (as of December 2025). This assumes that merchants directly and accurately organize their data. (TechCrunch)

Ultimately, accurate product data remains crucial. Information such as product name, model name, price, inventory, shipping, reviews, return conditions, comparison criteria, intended use, pros and cons needs to be structured. Data that is easy for AI to read is also useful for humans.

The problem is this.

The possibility that GEO will be needed and the claim that ordinary people can make money with GEO are completely different things.

Some marketing phrases mix these two.

Here's how it goes.

• Shopping exposure spots on ChatGPT, Perplexity, and Google AI Mode are mostly already integrated with merchant catalogs and payment infrastructure (Shopify, PayPal, Stripe). In other words, "how to be well exposed in AI shopping" is practically a matter of merchant data hygiene, payment integration, and catalog synchronization, not a clear space for external GEO consultants to intervene like SEO.

• Sales attribution measurement is also ambiguous. There are no standard metrics yet to measure outside the seller whether a product recommended on ChatGPT actually led to a purchase.

My conclusion is this: GEO is not currently a marketing channel for ordinary people to make money, but rather an infrastructure improvement where merchants and platforms align their data.

Who Actually Makes Money in the AI Age

We need to look at this objectively.

While it's called the AI age, there are still limited cases of ordinary people making stable income from AI. The publicly visible revenue structures are generally as follows:

Even LLM operators have large revenue, but they incur significant infrastructure costs and R&D expenses.

• According to The Information, OpenAI generated approximately $43 billion in revenue in the first half of 2025, while R&D expenses were $67 billion and cash consumption was $25 billion during the same period. Reuters stated that it did not directly verify these figures. (Reuters / Investing.com)

• Subsequently, OpenAI CFO Sarah Friar confirmed approximately $20 billion in total revenue for 2025 and reported that weekly active users had increased to 910 million. However, the same data also mentioned estimated inference costs of $84 billion in 2025 and $141 billion in 2026. (Sacra)

In other words, while revenue is growing rapidly, the structure of AI service operations means that inference costs also increase exponentially as revenue grows. Therefore, the notion that "AI services = automatic money-making machines" is inaccurate.

Currently, it seems that those who "sell how to make money with AI" are making money first. This includes lectures, YouTube, consulting, automation agents, prompt templates, and side hustle content. Not all of these are bad. There are indeed helpful educational resources and tools that increase productivity. However, we should be cautious about statements like "making automatic income just by using AI."

Why I Got Involved with GEO

My interest wasn't in simple blogs or marketing articles. I was interested in used prices, product comparisons, price tracking, product databases, review summaries, and purchase decision assistance.

So, the conclusion is this:

The concept is right, but the timing for ordinary individuals to make money with GEO is still too early.

Someday, AI will be able to read, compare, recommend, and even connect products for purchase. This is already happening in some categories. However, I don't think there are enough stable opportunities for external GEO business owners to make a profit on top of that.

So how should ordinary people view AI?

I don't see AI negatively. In fact, I consider it a very useful tool.

Document summarization, code writing, planning organization, data research, translation, idea review, and automation of repetitive tasks already have real value. I have also changed many of my work methods by using AI.

However, usefulness and profitability are different things.

AI is already a reality as a tool. However, the market where people make money with AI still has a large gap between "where big revenue flows" and "where ordinary people can enter."

Conclusion

I think the attitude needed now is neither overheated nor cynical.

AI is undoubtedly a big change. The market and revenue are growing rapidly. However, this change does not immediately translate into "automatic income for ordinary people" or "a new gold mine in marketing."

The same goes for GEO. It's true that AI searches mention brands and products, ChatGPT processes payments, and Perplexity handles real transactions with PayPal in the US. But these positions are mostly sandwiched between merchants and payment/catalog infrastructure, and standardized slots for external GEO businesses to enter haven't been created yet.

Especially in shopping, fast searching, price comparison, shipping conditions, reviews, coupons, and payment convenience are important. AI shopping is gaining a foothold in categories where comparison is heavy, but it is difficult for chatbots to immediately replace online stores in all categories.

The oo.ai case also seems to have tried to break through this point. The possibility of fast AI search, caching, indexing, and advertising integration models existed. However, before the service was sufficiently verified in the market, it collapsed due to representative risks and funding controversies, eventually leading to the domain sale stage around April 2026. This case is less evidence that "AI search makes money" and more of an example showing how heavy and vulnerable to governance risks AI search businesses are.

So, I would like to summarize it like this:

The most reliable way for ordinary people to make money in the AI era is not "a structure where AI automatically makes money" but using AI as a tool to solve existing problems better.

The same goes for GEO.

"Tricks that look good to AI" are not the essence. Building good product information, accurate price information, reliable comparison information, and verifiable review data that both humans and AI can read is crucial.

However, whether this is already a market where anyone can make money is a separate issue. I still don't think so.