How to get recommended by ChatGPT, and why AI picks the brands it does

For ChatGPT or any AI assistant to recommend your brand, your content has to give it a convincing reason why your product fits the situation the person asked about. That is how Seyong Park, founder and CEO of the Korean search data company Ascent AI, put it on episode 23.

It has to be convincing enough why our product is the right fit for that situation.

Seyong Park

When people talk about getting recommended by ChatGPT, the conversation often starts with technical files such as llms.txt. Park says to start with what is written on your own website. This post covers the part of episode 23 about AI recommendations and content (5:43 to 12:40 and 20:35 to 27:32). The episode is in Korean. Every quote below is translated from Korean, and each section links to its moment in the video. The original Korean quotes are in the Korean version of this post.

✦SEO and GEO are up to 70% the same work

Watch this part of the episode 01:59

The episode opened with the question from episode 17: is GEO the same job as SEO? Park answered from a marketing point of view.

If you look at SEO and GEO not as special, separate techniques but as part of marketing, the problem the two of them solve is the same.

This is how he defined GEO.

GEO is making sure that, in the AI search that started with ChatGPT, your brand or product gets recommended, mentioned or cited favourably inside the answer.

Asked how much the two overlap, he first said 40 to 50%, then raised it to as much as 70%.

Up to around 70% of it is the same work underneath, and there are only the work differences that come from the characteristics of the search engines I mentioned earlier, that is, traditional search engines versus AI search engines. The basic goal and purpose are similar.

The difference between SEO and GEO is covered in more depth in the episode 17 post and the episode 1 post. This post stays on the next question: what do you need to write to end up inside an AI answer?

✦AI search starts from a situation

Watch this part of the episode 06:58

Park's book, published in Korean in 2026, is titled Brand, Speak the Language of AI (브랜드, AI의 언어로 말하라). I asked him about an idea from it. Search used to be a way to tell people how well known a brand is. In AI search, what matters is branding that solves a specific situation. He said he spent roughly the first two chapters of the book on it, and he starts from the moment someone writes a prompt.

The moment we write a prompt is, one way or another, the moment we want to buy something, need something, or have a problem.

That moment is a buying situation, a situation with a worry, or a problem that needs solving. For your brand to be the one recommended in that moment, the condition is the one quoted at the top: it has to be convincing why your product fits that situation.

In his view, older search engines judged text in simple ways, such as whether the title contains the keyword or whether an important query is repeated somewhere visible, so their resolution was low. He sees AI as different.

But now AI behaves as if it understands our language really well, and the resolution has gone way up. So what I wanted to say is that the most important place to start is to keep saying, consistently and well, day to day: this is a brand, this is a product that can really solve that problem.

✦Why content comes before technical fixes

Watch this part of the episode 08:51

I also asked why he did not open with technical topics. He had just mentioned how many people now say you only need to turn your site into Markdown files or add llms.txt.

People say that if you put llms.txt next to robots.txt, AI will read all your pages and you will show up. It is not a story I had never heard before, but it felt like it was misleading people about what really matters.

He said the view comes from long experience with SEO.

I have done SEO in Korea since 2013, and before that, in Japan, I had been driving real traffic with SEO since 2006 for our own company's service.

Over a long time I learned that, in the end, there is no growth without the power of content.

What worries him is a brand that starts GEO with technical settings alone and is then disappointed.

Then two or three months later they have done all the GEO those people talked about, and nothing much has changed.

If that happens, he thinks the current attention on SEO and GEO could disappear, so he had a request for people working in the field.

Rather than talking as if there were an easy answer, people in this field need to spend their time and effort solving the real, underlying problems one by one. Only then can our industry keep going.

✦Put the reasons on the website you already have

Watch this part of the episode 21:18

People ask AI about a huge range of situations. I asked whether brands should write answers to all of them on their websites. Park started from the fact that brand channels existed before AI did.

It is not that you have to do something new, something completely different, to show up. You make good use of what you already run: your brand website, your D2C store, the stores you run on outside platforms, your blog, the page where you collect your press releases.

The channels are already there. What has to change is what goes into them.

But what you put inside them has to contain, from the AI's point of view, a reason to recommend this brand to people who are dealing with that situation.

He applied the same test to the idea that sending out press releases or posting in outside communities is enough.

People say AI puts a lot of weight on evidence and testimony from outside earned media, so that gets you visibility. Yes, that is definitely real. But what you write there is what matters.

✦The running tumbler example

Watch this part of the episode 22:00

Park gave one example. Someone asks an AI assistant this.

"I want to buy a tumbler. I've started working out, so I'd like one that's good for running, good to use while exercising."

The AI reads this as someone using a tumbler while running and works through the conditions. If the person runs on a treadmill at the gym, the tumbler has to sit somewhere on the treadmill.

"I shouldn't recommend one with a base that is too wide."

And because the person drinks while bouncing around, it cannot leak.

"It needs to be a tumbler that doesn't leak."

So the content needs more than a spec sheet. It needs to say which situations each spec helps in.

These specs, and the story of which situations they help in, have to exist inside your content. It has to make sense when a person reads it, and when AI reads it, it has to go "yes, that's right." Only then can it recommend you.

✦Why "No. 1 in the industry" falls flat with AI

Watch this part of the episode 23:38

On the other hand, the pretty words and strong claims written to hook people lose their force in front of AI.

Number one in sales.

Claims like that are completely neutralised in front of AI.

Instead of claiming this tumbler is "good for running," you need content that explains why it is good for running by connecting it to the actual basic performance of the product you have.

He called this the very first place to start.

The attributes that can serve as a reason for AI to recommend you.

And saying those reasons fully through your own media, I think that is the very first starting point.

✦Entities, in plain words

Watch this part of the episode 25:37

When the SEO industry explains these reasons, it often uses the word entity. Park thinks the word itself gets in the way for brand and marketing teams.

The word "entity" itself feels like a burden.

Some viewers may not know the term, so I asked him for a short explanation. He used the tumbler again.

The situation, running, and then the gym where the running happens. These nouns exist, and the relationships between those nouns exist.

Basically an entity is a real thing. If you break entities down a bit more, your brand is an entity, but the situation in which the product or brand is used is also an entity, and if people use it for a certain function, that function can be an entity too.

Mapped onto the tumbler example, his explanation breaks down like this.

• Brand: the company that makes the tumbler.
• Situation: running.
• Place: the gym where the running happens.
• Function: the feature people use the tumbler for.

That is why a line like this gives AI almost nothing to work with.

Lines like "It sold a lot, number one, first in the industry," from the AI's point of view, are sentences where the entities it is looking for barely appear.

In my book I tried to work out how to explain this in everyday language.

✦Three things to check on your own pages

Held up against your own website, Park's points come down to three checks for today.

1. Does the page say which situations your product or service fits?
2. Instead of "great" or "number one," does it state the specs together with the situations they help in?
3. Before building a new channel, are those reasons on the website, store, blog and press page you already run?

The rest of episode 23 has its own two posts. How to choose the prompts to manage, starting from category entry points, is in Category entry points in AI search: how to choose the prompts to manage. Naver versus Google in Korea is in Seyong Park on Naver market share in Korea: 51 to 49 with Google. What I check for AI search on a Korean SEO project is under AI search on the Korean SEO page. You can also click a time in the timeline above to jump to it in the video.

✦Common questions