Understanding and optimizing for various large language models (LLMs)

AI-driven search engines are reshaping how users interact with content. Continuing our video series on generative engine optimization (GEO), Josh Lohr, Senior SEO Manager at Contentful, shared insights with Solutions Strategist Allan White on how businesses can adapt their content strategies to align with GEO while holding onto some tried-and-true practices. 

Video: Watch AI Search Series Episode 2

Content strategy in the age of generative AI

Generative AI Search Optimization (GEO) changes the search game we’ve all been playing. That's the takeaway from our first video, What is GEO and how does it differ from SEO?

Rather than surfacing a list of links, these systems generate direct, synthesized answers. That shift fundamentally changes what it means to be visible in search results — and it requires a corresponding evolution in content strategy.

“There's so much to keep up with,” Josh explained. “New large language models are being released on a regular basis, and they're evolving. So, naturally, we need to know how to optimize for each of these different LLMs.”

How integrated content strategies feed hungry LLMs

Historically, question-and-answer content has been owned by support or documentation teams, while marketing has focused on brand storytelling. But in the GEO era, that division of labor is a liability.

To stay relevant in generative search, organizations need a unified content strategy that bridges teams. The content that LLMs draw from needs to be discoverable, structured, and aligned with how real users phrase their questions.

In these terms, your organization would benefit from regular syncs between marketing, support, and product content teams. Collaboration ensures consistency and helps surface answers that reflect real customer language and intent.

Key differences among popular LLMs for search

Understanding the inner workings of different large language models is critical to optimizing for them. Referring to the latest research from SEER Interactive, Josh outlined three types of models and how they influence content strategy:

• Training data-only models use static, pre-collected datasets. They're prone to outdated information — "You might ask who is the monarch of the United Kingdom, and you’ll get Queen Elizabeth still," Josh noted. Visibility here relies on brand mentions and contextual associations, which can be influenced by traditional PR and user-generated content. “Increasing our brand ‘mentions’ is the leading tactic to improve visibility in these," he added.

• Search-only models rely on real-time web crawling. Traditional SEO tactics are highly effective here. Josh pointed out, "Search data models cite their sources ... and they appear to be here to stay, which is fantastic." Knowing how different platforms work is essential. ChatGPT, for example, pulls from Bing for live search results. That puts a renewed spotlight on performance in Bing — a search engine many have previously overlooked.

• Hybrid models blend the two. They typically default to training data but use search to supplement information on "flip queries" that are fresh, local, or highly specific. "They will look to the pretrained model for most queries, but look to the web for what we call flip queries," Josh explained.

(It’s worth mentioning that the table below is already out of date since Claude now uses search data, and thankfully cites their sources.)

Source: Seer Interactive

Public relations: New value in the GEO era

Public relations isn’t just about awareness anymore — it’s about visibility in LLMs. "There’s a renewed emphasis on PR that we’re seeing with GEO which marketers need to be paying attention to," Josh said.

LLMs often rely on training data that includes brand mentions and media signals. As Josh put it, "There’s growing confidence they use mention and word-pairing signals in their training data that can be influenced by traditional PR and content amplification." Traditional media placements, press coverage, and even user-generated content can shape how your brand is interpreted by AI systems.

A practical tip would be to collaborate with your PR team on keyword-rich press releases and ensure media mentions link to authoritative landing pages. It’s one of the most efficient ways to shape your brand’s presence in training data.

How to future-proof your content strategy

Josh offered a clear roadmap for content teams navigating this shift:

• Build for answers, not just keywords. Focus on creating content that answers those questions directly and helpfully.

• Focus on brand visibility, not just page rankings. In training-data models, mentions matter more than metadata.

• Adapt tactics across models. This knowledge alone is critical to the optimization tactics we use in order to effectively be visible in each LLM.

And while staying current is important, not everything needs to change. "There’s so much to keep up with ... but much of your current SEO strategy is applicable to GEO," he said.

Building your business on a robust, API-first content management platform lets you recapitalize your investment in composable content. With flexible and powerful content modeling capabilities like Contentful’s, connecting your structured question-and-answer content from your support team with your existing public-facing content is easier than ever. 

This improves your content’s optimization for LLMs and traditional crawlers, so your brand will be surfaced among AI-generated responses when your customers are searching. 

"There’s a renewed emphasis on PR that we’re seeing with GEO which marketers need to be paying attention to."

Up next: Three playbooks for GEO readiness

Josh helped us understand how content strategy needs to adapt in the GEO era — but also what parts of your current strategy you can keep. Surprisingly, some relatively older approaches like traditional PR shine here. 

Our next episode will be bursting with details and depth on these subjects. Josh will take us down the rabbit hole of the details we need to understand in order to succeed in the GEO era of content strategy. 

We'll also be talking about knowledge graphs — structured databases of facts and relationships that help LLMs connect entities across queries. They underpin how generative engines understand context, and they play a growing role in content visibility. You’ll learn how knowledge graphs set the stage for AI search and content creation, and why cross-functional collaboration in your company is more essential than ever to provide those answers the search engines are expecting. 

"These aren’t optional anymore," Josh said. "If you want to earn visibility in generative results, you need to start thinking this way."

We'll also look at new ways to handle your media, and finally, the biggest reason to tune in: Josh shares his three playbooks for GEO, mastering authority, technical SEO, and the content playbook. Don't miss the next episode of our GEO Search Series, coming soon.

Ready to evolve your content strategy for the AI search era? Contact our sales team to learn how Contentful supports GEO-ready content delivery. And don't miss the first video in our series: What is GEO and how does it differ from SEO?

Take a tour

Chat with our team

We'll show you how our composable content platform can enhance 1:1 customer connections and drive growth for your business.

Subscribe for updates

Build better digital experiences with Contentful updates direct to your inbox.