LLM search optimization FAQ

How we find information, how we research things, how we search is changing.

LLMs initiated a shift from constrained, search box queries that return lists of links to natural language questions that return conversational answers — and it’s hard to imagine going back.

It’s an exciting shift, but it’s also a bit jarring and ambiguous, too.

We built this FAQ to be an ongoing, updated resource for our clients and fellow marketers to stay ahead of the latest updates to LLM search, LLM optimization, and all things Google search as we learn them.

Have a question that’s not answered here? Reach out and we’ll answer it in this FAQ.

Am I losing traffic to AI overviews? What can I do?

If you were building search-optimized content prior to AI Overviews being rolled out, almost definitely you will see a decline in clicks for any top-of-funnel, informational content.

Google's AI Overviews provide comprehensive answers directly ahead of standard search results, reducing the need for users to click through to websites for informational queries.

Reports indicate that clickthrough rates have decreased 30%+ due to the introduction of AI Overviews.

We are encouraging teams to spend their strategic writing resources on content that brings forward their unique perspective and point-of-view and to write content as context—context that they’re uniquely qualified to write. This context can then be repurposed into various content formats through LLM-assisted workflows to maximize its value.

What is Google AI mode, and is it the new way to search Google?

AI mode is Google’s latest innovation on their search experience. It’s an LLM-first search experience that feels more like what users are used to from ChatGPT’s web search functionality than the traditional Google search experience:

AI mode was introduced in March 2025 and is now rolled out to all US users. To access, it, you can go to google.com/aimode and follow the appropriate instructions to enable it for your personal Google account. It’s not clear if this will become the dominant search experience, but we think some version of this is likely.

We’re closely following the rollout and response to AI mode and will keep clients and this FAQ updated we as learn and observe more.

Why would I use Google AI mode over ChatGPT?

ChatGPT requires you to “search the web” specifically, or otherwise requires you to prompt it to search the web, whereas Google AI Mode will search the web 100% of the time, no questions asked, using a query fan-out approach. Want the most recent data available on the internet? For now, AI mode would be your pick. For a deep dive on how AI mode works and the future of search, How AI Mode Works by Mike King is one of our latest favorites.

When do LLMs search the web vs use training data?

Every change to the state of your query for an LLM can have a significant impact, and it’s important to understand what’s happening.

As of May 2025, every major public LLM provider has web search integrated. There are important differences to know in how web search is triggered, which are not fully understood, depending on the LLM provider.

At the highest level, for most LLM chat surfaces, if the LLM has sufficient training data for a query up to it’s knowledge cut-off date, it’s unlikely to trigger a web search.

Currently how this works depends on the provider, and the level of control and visibility is changing often.

As a way to “force” web search more reliably, you can prompt the LLM to “Search the web for latest info as of today’s date.”

• ChatGPT (GPT-4o): Appears to utilize live web search when training data is incomplete. If ChatGPT returns citations with its response, its a strong indicator it searched the web. You can also trigger a web search using the Tools dropdown within ChatGPT by selecting search the web:
• Claude 4 (Opus and Sonnet): Offers web browsing capabilities but searches the web under specific conditions. A recent Claude system prompt leak gives more detail into how it works: “Use web_search only when information is beyond the knowledge cutoff, the topic is rapidly changing, or the query requires real-time data.” To trigger web search reliably, prompt Claude to search the web.
• Google Gemini: Able to search the web, with no specific option to do so.
• Perplexity AI: Always-on live web search with cited sources
• Google AI Mode (beta): Always searches the web with cited sources using a query fan-out approach

Do LLMs credit sources when using my content?

Often - yes! But also, no. ChatGPT, Claude, Gemini, and Perplexity provide citations or links to original source when they’ve searched the web. Unfortunately, when they retrieve information from their training data, this is often unavailable or unreliable.

In fact, if an LLM doesn't include citations in its initial response, this is a strong indicator it relied on its existing knowledge without triggering a web search. If you later ask what sources it used for its response, there's a good chance you'll hit a broken link and/or that the citations are a best guess or hallucination. This is because LLMs are not designed for source retrieval and can't accurately cite sources.

How do I optimize my content for LLMs?

There is no “rank” or “indexing” for LLMs. And even within specific industries, we’re seeing research indicate that LLMs favorite sources to cite (eg Reddit, Quora) are rapidly changing.

For the popular general-purpose LLMs, we have little visibility into how they work, but can track citations over time to understand patterns in behavior. However, it’s important to note that this is currently being heavily researched with emerging learnings regularly being surfaced. Josh Blyskal at Profound has been regularly posting interesting research on this topic.

Whether or not you show up in an LLM query is incredibly dynamic—possibly even moreso than search, and certainly much more volatile at this stage. We track this information closely with clients across popular public LLMs and are regularly performing analysis of the data.

That said, here are some things we can at least semi-confidently recommend for optimal visibility:

• Identify blog post author and their role to establish authority
• Use authoritative sources in your copy and cite your sources
• Lead with active verbs and use short sentences so content is easy to parse
• Structure content like components that can function independently (eg: a paragraph should be able to live alone without missing key details)
• Add FAQs at the end of post to capture potential queries a user might ask about the topic
• Leave at least 15% of any blog post for connection to product. This makes your content inevitably unique in the SERPs and gravitates your strategy towards content with strong connections to your product (= higher conversion)

Technical site concerns should also be reviewed, but are rarely an issue. We’re happy to help if you’re trying to sort through a confusing audit or need some guidance.

Can we influence what LLMs say about our brand?

Indirectly, yes. By ensuring that authoritative and up-to-date information about your brand is available across reputable sources—such as your website, Wikipedia, social media, and industry publications—you can influence the data LLMs access and, consequently, the narratives they generate.

What should we communicate to stakeholders about the impact of LLMs on our marketing strategy?

The rise of AI-generated answers is reshaping how users find things. A strong content strategy to maintain and increase visibility and authority is more important than ever.

In order to obtain optimal and accurate search ranking and LLM visibility, investments should be made in messaging and content refinement on and off property, creating net new content and repurposing at scale through a content systems approach, and refining website messaging and updates.

Now is the time rework KPIs and organization goals to align with the shift.

What’s the relationship between search engine optimization and LLM optimization?

The premise—as in “optimizing for how people find things”—is the same, but almost nothing else is.

SEO historically has helped brands go from unknown to known, across a variety of topics that they otherwise might not have been surfaced for—through content. You write content focused on topics and derivative keywords, and with enough focus, it can rank.

LLMs don’t follow the same rules—and it’s complicated.

But whether or not you’re surfaced by LLMs is still at least partially determined by SEO. Search engines like Google and Bing are a grounding factor for LLMs. This means existing rank might still help. Research is actively happening here as well.

On the bright side, organic search is still a strong channel for those who have invested. It’s not easy to see gains there right now, but investing in LLM search as a new channel has lots of potential upside.

If you’re investing in SEO today, we recommend continue everything you’re doing, but also have a complimentary strategy for LLM visibility—and stay on top of the latest developments, optimizations, and opportunities.

Turns out, you need to talk about your brand everywhere now. Authoritative sources are the new Google. Every search on Reddit, YouTube, or anywhere your users are is its own SERP, and you need to show up.

As for what content to create, our advice is to double down right now on a handful of strategic topics and use a content systems approach to obtain maximum coverage for each topic. With AI-generated content dominating many SERPs, we recommend standing out by building content that only you can create with your unique perspective and point of view.

Have a question that’s not answered here? Reach out and we’ll answer it.