Is ChatGPT leaving your brand off the shortlist? LLM Pulse founder Esteve Castells on how to win at GEO
Generative engine optimization (GEO) is the work of getting a brand cited and recommended in AI answers from ChatGPT, Gemini and Google AI Overviews. In a MarketScale webinar, LLM Pulse founder Esteve Castells said GEO builds on SEO but rewards citations over backlinks, decision-focused content over generic explainers, and a presence on the third-party sites that make up 96% of what AI reads. He recommends setting a prompt-tracking baseline, tracking about 20% brand and 80% non-brand prompts, and repeating experiments that earn citations.
This story was produced through MarketScale. See how Marketing Tech teams put it to work with AI Writing.
Key facts, context, and what it means.
Key takeaways
GEO uses the same ingredients as SEO, but citations from sources that talk about your brand count at least as much as backlinks.
LLM Pulse data shows 96% of what AI reads comes from third-party sources such as LinkedIn, YouTube, review sites and competitor pages.
Track about 20% brand and 80% non-brand prompts, and create content for the questions customers already ask your support team.
Free workspace
Turn your Marketing Tech expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
Picture your best-fit buyer on a random Tuesday night. They're stuck on exactly the problem your company solves. Five years ago they would have Googled it, scrolled a few blue links and maybe landed on your blog. Tonight they open ChatGPT or Perplexity, type the question and get one answer back: three vendors, a couple of sentences on each and a recommendation.
Were you on that list? And if you weren't, how would you know?
That question drove How to Succeed on GEO in the Era of AI Search, a MarketScale live webinar hosted by Daniel Litwin at MarketScale HQ in Dallas. His guest was Esteve Castells, founder of LLM Pulse, the technology behind the AI Visibility tool on MarketScale channels. Castells has spent more than 10 years in SEO, leading search work for some of the largest companies on Google, and now consults on GEO, or generative engine optimization.
Why AI search changed how B2B buyers shop
LLM Pulse itself started with a B2B consulting client, an HR marketplace software company that wanted to show up in ChatGPT back in January 2025. The turning point, Castells said, came when ChatGPT gained access to web search. Before that, models answered only from what they learned in training. Now they search the internet, read more sources than any person would and come back with three to five recommendations. "We have a new actor with this AI, which makes a lot of decisions for you on what to filter, what not to filter, and that is a paradigm shift," he said.
He put ChatGPT at about 1.2 billion weekly users, with Google close behind through AI Overviews and Gemini. For marketers, Litwin said, that means a buyer may build a shortlist before ever visiting your site. The website still matters, but more as a source of truth than as the first touch.
Is SEO dead? What changes with GEO
For marketers who spent a decade mastering SEO, Castells had reassuring news. "SEO and GEO have basically the same ingredients, but GEO cooks the ingredients differently," he said. Teams that did SEO well start with an advantage.
Technical SEO remains the foundation. Some of it even matters more, because AI crawlers are not as advanced as Google's and may miss content that only appears through JavaScript. What changes is the role of backlinks. Castells said brands now need to think in terms of citations: web sources of any kind, your own site or other people's, that talk about you. For AI, citations count at least as much as backlinks.
The field is also more crowded. Google used to be the only player that mattered. Now, depending on the market, a brand may need to show up in ChatGPT, Gemini, Claude, Grok and others.
What content gets cited by AI
The content most at risk, Castells said, is purely informational. A post explaining what a coffee machine is answers a question AI can handle without searching at all. Content that informs a business decision is a different story. When AI researches coffee machines, it looks up reviews of every model, user opinions, alternatives and prices. If a brand doesn't publish that information, AI takes it from a third party, even an outdated one.
"Some people say content is not so relevant, but they are completely wrong," Castells said. "Content is more important than ever."
That makes comparisons, alternatives pages, pricing and listicles valuable. Litwin pointed to music software makers that publish neutral comparisons of their product against rivals. AI isn't looking for your most persuasive pitch. It's looking for sources of truth, and a fair, useful comparison gives it a reason to cite you.
Castells also urged brands to make specific claims. Instead of calling yourself cheaper, list your price next to the competitor's so AI can map what each brand offers. And make pages readable for machines. A pricing table built from green ticks and red Xs may mean nothing to a bot, so add "yes" or "no" in plain text.
Why 96% of what AI reads isn't your website
According to LLM Pulse data Castells shared, 96% of what an AI reads comes from third-party sources: competitor websites, YouTube, Reddit, LinkedIn, review sites like Trustpilot and other user-generated content. A brand's own website is a small slice of the picture.
That changes how reputation works. In classic search, buyers rarely went digging for reviews. AI reads every opinion and summarizes it, so a brand no longer controls how it's described. Castells said negative sentiment can hurt more than not being mentioned at all. Litwin's takeaway was that expertise needs to come from many voices, including engineers, salespeople and customers, not just one brand-approved spokesperson.
An audience member asked whether publishing the same piece on several platforms helps AI visibility. Castells said yes, and that brands should do it on purpose, because different AIs lean on different sources. YouTube is powerful across Google's products but, he said, close to invisible in ChatGPT, where LinkedIn has become the second-largest source. Format matters too. Shorts often outperform long-form YouTube for the brands LLM Pulse works with. On LinkedIn, polls, articles and posts with video do well. Instagram, Shorts and TikTok videos are also getting pulled in as sources.
Employee sharing helps for a similar reason. AI understands when a business is posting about itself, Castells said, and tends to look for more genuine opinions from real people. Litwin added that most companies already have material to work with, such as old podcast episodes or blog posts that can be cut into short clips and reshared by staff.
Is Reddit still worth it for AI visibility?
Sources shift fast. A month earlier, Castells said, he would have told brands to focus on YouTube and Reddit. ChatGPT updates have since cut Reddit's weight sharply. Asked why AI trusted Reddit in the first place, he said it was one of the few places with huge amounts of human-written content, but it now carries a lot of bots, incentivized posts and scandals involving subreddit moderators. He doesn't recommend most brands spend time there. His three pillars instead are a brand's own website, social channels and user content, and earned media, which he said is making a comeback.
How to write for the searches AI runs
Another audience question asked whether content should be built around the questions buyers type into ChatGPT. Castells pointed to query fan-out. AI breaks one question into many searches. Ask for the best laptops for work and it may search for the best-rated laptops of 2026, then reviews of individual brands and models. Those searches drive citations and mentions. "It's not so much that you need to think of, hey, what is the user searching," he said. "Is it more like, what is AI searching to satisfy the user?"
The best raw material is often already inside the company. When clients ask him for the perfect prompts, Castells said, he asks for the top 10 questions customers raise with the support team, and has them create content for those first.
A four-step GEO checklist for B2B marketers
Asked by an audience member for a starting checklist, Castells laid out four steps.
- Set a baseline. Track prompts before you change anything, so you can tell whether your work paid off.
- Analyze that data for quick wins, mixing tool insights with your own.
- Optimize in a loop of experiment, measure and roll out, with citations as the priority. Being cited leads to being mentioned, and a mention means AI is recommending you.
- Learn which channels matter for your category. Some large B2B categories have almost no Reddit discussion, so effort there would be wasted.
The most common mistake he sees is tracking the wrong prompts. Plain informational prompts tell you little, and prompts that name your own brand inflate your numbers until you look like the most mentioned brand in the world when a competitor may be five times more visible. His recommended mix is about 20% brand prompts, including reviews, opinions and alternatives, and 80% non-brand prompts that mirror real business conversations, such as the best laptop for work on a $1,000 budget. He also warned against chasing trends like llms.txt files before the basics are in place.
Above all, he said, act. He recalled a client that studied which media outlets had content deals with OpenAI before a PR push and then ended up doing nothing. Creating content beats overthinking it.
What comes next: always-on AI agents
Looking a year or two out, Castells pointed to always-on AI agents that work for people around the clock. A travel company, for example, may soon compete against agents that check prices 24/7 on a user's behalf. He expects professional use, and B2B software in particular, to lead adoption. "It's very early days," he said. "Still many companies are not even investing in GEO."
Castells can be reached on LinkedIn at linkedin.com/in/estevecastells and by email at esteve@llmpulse.ai. LLM Pulse is on LinkedIn at linkedin.com/company/llm-pulse. MarketScale customers can track their own visibility in the AI Visibility tool on their MarketScale channel.
Sources
- Esteve Castells on LinkedIn ↗ · LinkedIn
- LLM Pulse on LinkedIn ↗ · LinkedIn
- LLM Pulse ↗ · LLM Pulse
Featured companies
Your experts belong here
Every story in MarketScale Marketing Tech starts with a company putting its practitioners, product marketers, and RevOps leads on the record. Buyers are already reading this topic. The only question is whose experts they find.
Your buyers live in search and AI answers, so published expert content is the channel that compounds instead of expiring.
About the author
Daniel Litwin is a journalist of multiple disciplines focused on finding and telling engaging stories for B2B communities. He has interviewed executives from Fortune 500 companies including Honeywell, Microsoft, John Deere, and Chipotle, and leads editorial direction at MarketScale. Litwin hosts weekly shows and podcasts while helping develop new content approaches across the MarketScale platform. He holds a B.J. in Radio/Television Reporting/Anchoring and a B.A. in Spanish from the University of Missouri-Columbia.
