Roman says: “Everyone should be doing GEO or Generative Engines Optimization. It's not that SEO is dead; it has changed.”

Okay, everyone should be doing it.

Why should everyone be doing it if the vast majority of the traffic still comes through traditional search engines at the moment?

“Simply put, as you might know, Surfer is a content SEO/content optimization platform. I'm not going to lie, as of today, we get almost as much organic traffic from all LLMs combined as we get from Google SERPs, which we have been writing content for for almost seven years, and our blog post ranks number one for many SEO-related keywords, like ‘content optimization’, ‘SEO’, et cetera, et cetera.

Now we see that ChatGPT, Perplexity, Gemini, and AI overviews combined make up almost as much as Google search brings us, or the classic slide of the old search.”

But is that not likely to be unique to your particular industry/your particular product?

“Absolutely not, because LLM search, and LLMs usage as a new tool or as a new product, has only kept growing. ChatGPT has gained 700 million unique users a month in less than a year since it launched.

It's been more than three years since they started. Now there are more competitors. So, its share is growing and will only keep on growing. If a company like ours already has as many people coming to our website from LLMs as from Google, who had almost 30 years to get their ground, I think this change will come much faster than it took Google to gain ground.

Also, the Google product manager, behind Gemini and AI overviews in particular, has posted on Twitter just a few days ago that they're planning to make AI Mode a default likely by the end of this year. I'm speculating on the date, but he claimed it would be in the nearest future, so soon it will be LLM search everywhere, including Google itself.

I'm not touching the smaller players like Bing, DuckDuckGo, etc., but Google itself is likely to switch fully to AI Mode, so it would be LLMs. Whatever you look for would be read by an LLM before giving you any output, so you would not get involved in classic search anymore in the near future.”

So, what you're essentially saying is that you're seeing a lot of traffic at the moment. It may not necessarily be reflective of every industry, but what you're seeing is exponential growth, and if you simply follow the curve, you can see that within a couple of years, many industries will be the same as well.

“I think it will become a default for every industry, as nowadays you can argue that some queries sit better for LLM search. Some queries still fit better for the classic search.

For instance, if you look for something like an informational type of query or informational intent, you might still default to the classic search to open a page on a website you trust – for instance, Wikipedia – and go through the article yourself and try to find all the nitty-gritty details which you are interested in about a particular topic, because it's an informational type of thing.

But if it’s a commercial type of query, for instance, which is probably the most crowded type, and has always been the most competitive part of search – whether it be advertisement or optimizing for ranking for a particular commercial intent type of query. But you can already see that commercial queries are a much better fit for LLMs for search, because it would already go through a bunch of sources and compare what people are saying about a particular product you're willing to buy.

For instance, if I Google something like, ‘best air fryers under $200 to buy 2025’, I would default to go with this question (personally, I mean, I'm speculating) to ChatGPT or Gemini right away, instead of going through 10 comparison pages Google gave me and spending 20 minutes on that.”

Okay, so you mentioned ChatGPT and Gemini, obviously, essentially powering Google AI Mode there as well. Why are you focussing in on those LLMs?

“ChatGPT is 75% of the market share right now. As I said, Google makes up for maybe 10-15% (AI overviews in particular, not Gemini though. It's a fraction of a percent maybe – or like 1%).

But as I said, ChatGPT is the king because they were the first and it's the most popular LLM. It’s not just LLM's first search, it's just the most popular LLM that happens to have the search mode as well.

Google's still the king there in terms of AI overviews, and again, as I said, they are likely to make AI Mode default. I do see the future where it's going to happen one way or another soon.”

What is the process for optimizing for LLMs, and how is that different, if at all, from optimizing for more conventional search?

“The process is still in the making. It's something so new, and it's being changed as we speak, probably. I can only talk about something that wouldn't change at least by the end of this year, maybe. What we see our customers are doing right now (huge companies with dedicated SEO or marketing teams) and what we do internally, and what we're trying to now build into our products to automate this process.

You start by coming up with a bunch of prompts you would like to rank for in either LLMs or AI overviews, Gemini, etc. Let's say 15. If you check and find that you do not track for some of those prompts, you start tracking them. There are plenty of AI tracker tools on the market, including ours. They work relatively similarly. I don't think there's a huge difference between which AI tracker tool you will pick.

What makes a difference is how you would then either re-optimize your existing content to increase the chances of being picked up as a source or cited by LLMs, or write new (as I call it) GEO-first content.

As an example, what we do internally, we pick up a bunch of prompts. For instance, it might be something like, ‘Surfer versus Competitor A’, and we see that we are not even being picked up as a source or not cited in ChatGPT for this type of query. What we do next is we go and find all the blog post pages we have related to the topic, the old ones, and we include the so-called ‘facts’.

We call them facts, but they are basically statements. It's not that they're backed by something. By ‘fact’, I mean a 50- to 100-word sentence or a short paragraph that makes a statement. For instance, ‘Surfer costs $99 a month for X type of plan.’ That's a fact. Then, we integrate a piece of those factual claims within our existing blog post pages that are semantically relevant to a search prompt or a query we would like to rank on those LLMs, and we are currently not being cited.

The second activity we do, for this particular case (Surfer versus a competitor), is we would put up a comparison page, which would only answer the questions or bullet points that the LLM response had. If you ask, ‘What are the advantages of Surfer compared to competitor X?’ and it returns you a bunch of bullet points: ‘It's better for feature one, feature two, feature three,’ we would write a page which would specifically address those feature one, feature two, and feature three things.

That's something that anyone can do, and it would work for pretty much any industry, any company size, etc.”

A lot of SEOs would push back and say you have to have human-first content; you have to begin with authority, experience, and trust, and demonstrate that, show that it's a real person, and perhaps base your article on some kind of video or podcast that the AI can understand that it's humans who have produced it first.

What would you say in response to that?

“I don't see how it's supposed to conflict, by including those facts and tweaking the structure of your content just a bit.

For instance, our correlation study also shows that AI or LLM search prefers a specific or certain structure in the content, like bullet points, tables, lists, etc. Just because it's easier for the LLMs to read.

By turning some of your human-first written article from the first person point of view (if you've been on a trip to something and you have the travel blog and you're describing how did you like your time spent in Rome and giving suggestions to people), if you turn a bunch of this content into a table, it wouldn't change the way it reads to actual people who read it if someone ends up reading your whole blog post on your website.

It's an additive thing. Optimizing pages for LLMs doesn't conflict with optimizing for SEO, as such. These are like two sides of the same thing. They're more additive than conflicting. You just pre-call a bunch of stuff on top of the existing SEO piece/SEO blog posts, and they don’t hurt each other.”

You've talked about maybe optimizing for LLMs first, or primarily LLMs. Would you advocate publishing articles that are just targeted at LLMs and publishing other articles on your website that are targeting real humans?

“That's a great question. That's case by case.

The example I provided, where, for instance, if we at Surfer find that we do not rank for a particular prompt in a particular LLM, we might (and it will be faster to) just write and publish the page, which would be particularly aimed. This page's purpose would be only to get us cited and sourced for this particular prompt in this particular LLM. SEO is almost like a byproduct here.

But if we have time and there is no urgency, we can incorporate the process within our internal SEO content writing process. When we have a brief for a particular article to write, an SEO article, we add features of LLM optimization that this article needs to have: put a table here, make sure to include those facts because they can’t come from LLMs, etc.

We already have this process pretty much well automated inside the platform. It's already a publicly available feature. It's available to users. We have the so-called ‘LLM facts’ available in all our content editors. That's the part we automated.

As I said, AI Tracker, we also have it, but there are plenty on the market. That doesn't make the cut by itself; optimization does. In terms of tracking, pretty much anything would do. You can even do so manually.”

It's interesting you mentioned a table a few times there as well, in terms of important elements to include within a blog post or an article.

What other elements or structural styles would you generally advocate a piece having in order to be more likely to appeal to LMMs?

“The one which has the highest correlation is as simple or as stupid as numbered headings. Literally, in your H1, you can label it as ‘Step 1:’ and your H1 (as it used to be) and then just ‘Step 2’ would be your H2, etc.

The thing to mention here is that Google has its crawl budget. The less it would spend in terms of compute time to crawl your page and to get any meaningful information from it, the better it is for your website and this particular page's performance.

The amount of tokens LLM would spend to get any meaningful data that would serve a user's prompt or query is the same as the crawl budget that Google has. So, the fewer tokens it will spend to find a particular piece of information the user looks for, the better it is for you.

Having a certain structure helps it because all LLMs work pretty much the same. It's basically transformer architecture with some different flavours or tweaks, but it's basically transformers. They can keep a context of the whole page at once, but they will still start to read from top to bottom: having the main query above-the-fold and having the structure, like number headings, tables, bullet points, etc.

You could say it reads every text as a human would. It's easier to skim through a bunch of bullet points than to read a wall of text and try to distil the information from there.”

One challenge with people consuming a lot of content on LLMs is that they do it on that platform and that they're probably less likely to click through to a website.

How do you measure success on LLMs?

“That's debatable, though, but it's not just our journal study. There's a fresh SE Ranking study that just came in, and they claim that people who came to your website from any type of LLM spent 68% more time on your website.

It sounds like common sense, though, because it pre-filters your intent. If you have already decided to open any website from an LLM search result output, it means that it did this pre-filtering for you. As in a classic search, when you get to the Google SERPs, and you go through a bunch of the first couple of pages, and then you land on the exact one you were looking for.

An LLM eliminates this part of the process. It gets you exactly the page you were looking for because it reads all of them on your behalf, basically.

Sorry, what was the second part of the question?”

Well, I guess it's in relation to traffic. I put to you that perhaps not as much traffic comes from LLMs to websites as traffic would come from more traditional search engines to websites.

You pushed back on that a little bit by also talking about the quality of traffic as well, but the main question really revolved around measuring success and what metrics you had in mind to measure whether or not the work that you were spending on trying to feature content on LLMs was worthwhile.

“It isn’t much different from the metrics you have and targets you have for SEO content, as such. It's basically unique users, people who came to your website, and people who completed the target action, whatever it is for you.

Are you selling the product or service? Are you just an informational type of website and you're getting paid for people who are seeing advertising on your page, etc? It's not any different. Luckily, LLMs like ChatGPT, for instance, have their UTM built in. It's easy to track, as well as pretty much any other LLM, so it's not a huge issue.

As I said, in terms of quality on this traffic, depending on your niche, category, or type of business, a user who came to your website from an LLM is much more likely to be a more engaged user than those who came from Google, just because of this intent pre-filtering has already been done for them and served to them.”

Roman, what's the key takeaway from the tip you shared today?

“I'm not sure it would be the key takeaway, but I also wanted to mention that mentions of your website or a particular page on your website and other websites are basically new backlinking.

To do well on GEO and to rank well on LLMs, LLMs prefer in-context natural mentions of your brand or product on different websites, rather than dofollow, 301, backlinks, etc. – as it works for classic SEO.

If you do not strike for a certain prompt in a certain LLM, but your competitor does or another website does, and you reach out to them saying, ‘Hey, can you please mention us in this way, mentioning these facts about us, and we will do the same for you on our website?’ it would do much better, or it would do as well as classical backlinking for the original classical SEO.”

Great, okay, and did you want to share something else in terms of a key takeaway?

“It's easier than it seems. It's going to be our new standard in the near future.

It's not going to become a separate industry or profession. I think it would just become naturally built into the existing SEO or marketing teams. GEO is just a new reality that everyone should start embracing, the sooner the better.

Start optimizing for the exact prompts you would like to rank in particular LLMs. Don’t just keep doing what you've been doing, writing the old or optimizing existing blog post pages SEO-first and pray for better. Do deliberately optimize for GEO as well.”

Roman Leliukh is AI Product Manager at Surfer. Find out more over at SurferSEO.com.