What to Know About AI-Driven Search
When it comes to AI discovery, nobody is an expert yet, but I have interesting findings that I would like to share. Here, I will walk you through the AI-driven search, how it’s changing the search landscape, how it’s different from traditional SEO, and look into the real-life case study – providing deep insights on what exactly we did, and the results we achieved. There’s a lot of excitement and, let’s be honest, a bit of uncertainty surrounding AI right now.
Many B2B projects are facing challenges, including decreased organic traffic and the impact of recent Google updates like the Helpful Content Update. For example, organic traffic is down between 15% to 30% YoY or even QoQ.
Clients are expressing frustration that simply ranking number one on Google isn’t enough anymore. The reality is, search has changed. AI is becoming ubiquitous, and traditional B2B SEO is no longer as effective.
The key concern? How do we continue growing traffic and leads in this new environment? Failing to optimize for AI risks losing reach, especially as platforms like GPT and Perplexity gain market share. Many are unclear on how to integrate AI into their marketing mix, leading to questions about what to do and why.
AI and LLMs vs. Traditional Search
Let’s address the core question: How do AI and LLMs differ from traditional search, and what steps can you take now as a B2B decision maker?
Traditional search starts with intent formulated as a keyword, aiming to provide a ranked list of relevant pages. LLMs, on the other hand, predict and generate the best next word in response to a query, operating as well-trained predictive machines.
Key differences include:
- Query handling: Traditional search is keyword-based, while LLMs consider context, intent, and broader correlations.
- Information retrieval: Traditional search retrieves indexed content, while LLM-driven search involves parsed content and synthesized answers.
- Result presentation: Traditional search provides a list of links, while LLMs offer direct conversational responses.
- Personalization: Traditional personalization relies on cookies and account history, while LLM personalization remembers conversation context.
- Knowledge Updates: LLMs initially had a knowledge cutoff, but now they have search functions and can learn about the real-time world. Traditional search always provides up-to-date information, provided that’s what ranks the highest.
- User Interaction: Traditional search involves short queries, while LLM interaction is dialog-based.
How to be LLM-Friendly: Key Principles
Despite the complexity, LLMs are still machines with algorithms, meaning they can be optimized. This is actually the good news – we’re dealing with machines that follow certain patterns, which means we can optimize for them.
Step 1. Get Inside the Source Code
LLMs are trained on massive datasets like Common Crawl, which serves as a foundation for their knowledge. They learn about the world from pre-training data, including those from around the web. Think of this as the “early days of Google” equivalent when you had to be registered in certain directories to be found.
How to check what Common Crawl knows about your business:
- Check if your brand is indexed by Common Crawl using their index checker tool
- Look at the most recent monthly crawl data (they update literally every month!)
- If you’re not there, this is your first priority – you need to be in that data set so machines know about your brand
- This is how you get right to the core, educating the LLMs about your business from the training stage itself
Step 2. Understand What AI Models Actually Know About Your Brand
Before improving your visibility across AI-powered platforms, it is essential to understand what large language models currently know or assume about your brand. This foundational step helps determine how AI systems present your business to potential customers, partners, and decision-makers.
Vertology is the purpose-built tool that provides this insight.
Developed without writing a single line of code, Vertology is a no-code application designed to audit how models like ChatGPT perceive your brand. It performs two complementary evaluations. The memory-based audit shows what the model retains from its internal training data. The live audit reveals how your brand appears in real-time queries using current web information.
Together, these audits highlight the alignment or disparity between your actual positioning and how AI systems describe you.
Many businesses discover that their AI visibility is outdated, incorrect, or misaligned. Vertology presents this gap clearly and provides practical recommendations to help close it.
The platform also enables direct comparison with competitors.
This feature reveals how your brand performs in AI search results relative to others in your industry. It is especially valuable for B2B organizations where differentiation often depends on clarity and precision in messaging.
Vertology delivers results in a structured and accessible format.
You receive detailed audit reports that include citations, visibility summaries, and strategic insights. These reports can be downloaded and shared across marketing, content, and executive teams to support informed decisions.
For companies shifting from traditional SEO to AI-driven discovery, Vertology offers clarity, control, and a measurable path forward. It answers the essential question: What do AI systems say about your brand, and how can you shape that message?
Step 3. Educate the Models About Your Brand
One of the most fascinating things I’ve discovered is that you can actually educate these models yourself through direct interaction.
What you can do right now:
- Build custom GPTs with your knowledge base (this is what I did, and now it remembers details about my business that aren’t even on my website!)
- Upload content about your business to platforms that feed into LLM knowledge
- Regularly interact with AI platforms to provide accurate information about your services
- The more you educate the platform about your brand, the more it knows – it does remember!
Step 4. Structure Your Content for LLM Comprehension
LLMs are still machines – they read your content and tokenize it, transforming it into numeric streams they can process.
What you can do right now:
- Use logical structure to make it easier for LLMs to process your content
- Double down on definitions, examples, and detailed descriptions
- Create content as if it was a ready-to-go answer in ChatGPT
- Work with your content to make the job of LLM models easier
- Remember: when you double down on these elements, you’ll appeal to both LLMs and humans
Step 5. Rethink Your Reference Strategy
LLMs rely heavily on references, but differently than traditional SEO approaches.
What you can do right now:
- Start referencing Reddit and Quora discussions with large followings
- Incorporate credible statistics and educational articles
- Change how you reference stuff compared to traditional methods
- This transformation of your content seems to improve how LLMs understand it
Step 6. Optimize Images with Context
If you’ve ever generated an image with AI tools, you’ll notice the file names are extremely long and descriptive – that’s a clue!
What you can do right now:
- Instead of short alt tags, provide more contextual descriptions
- LLMs can technically “see” images but rely more on words and context
- Give LLMs better context for what these images represent
- Think sentence-level descriptions rather than keyword phrases
Step 7. Implement Schema Markups
As they are machines, LLMs parse content similar to how Google does – so schema still matters!
What you can do right now:
- Add FAQs and schema markup to make it easier for LLMs to understand your content
- Implement HowTo articles and other schema.org structures
- Remember, it’s still a machine – when you give it structured data, it performs better
When you add all these elements together with the best SEO practices, this is where the results come in. We’ve seen this work incredibly fast – SEO has been put on steroids since 2022, and now with AI overviews, you can get even faster results. The game has changed, and these strategies will help you stay ahead!
The Blended Reality: Google and AI Overviews
Google has integrated AI overviews into its search results, appearing in more and more significant portions of searches. This represents a blend of traditional search and LLM-driven search. AI search is measurable, and tools like Ahrefs and Semrush can track which keywords and pages appear in AI overviews.
Case Study: Generative Engine Optimization (GEO)
We’ve seen great success with the Generative Engine Optimization (GEO) framework. In one case study with Case IQ, a case management platform, we focused on user intent, structured content, definitions, examples, detailed descriptions, explanations, links to glossaries, reviews, social proof, and referencing.
The results? This led to significant growth in keywords appearing in AI overviews in just one month, positively impacting impressions and targeted traffic. Below are the key results in only 2 months:
- +26% new keyword rankings in top 3 positions
- +33% increase in AI Overviews and featured snippets
- +426% more keywords in AI overviews specifically
Most interestingly, while measuring the results, we also noticed that the search volume for the keywords whose searches have AI Overviews increased from 3,330 to 16,750 (403%) in the same period.
Key Principles of GEO:
- Prioritizing user intent over keywords.
- Structuring content logically for LLMs.
- Providing definitions, examples, and detailed descriptions.
- Referencing reviews, social proof, and relevant discussions.
- Optimizing images with contextual descriptions.
- Adding schemas, FAQs, and how-to articles.
FAQs on GEO and LLM Visibility
How do you monetize LLM visibility?
From a business perspective, clients benefit from greater visibility in search volume. Getting more exposure on AI overviews leads to more traffic and exposure from Google. It’s a second revenue stream from Google. For pure AI models, clients are building custom GPTs to generate more traffic to their websites. People are building tools and assistants that help both models learn more about their business and send traffic to their websites. Think about it as the early stage where you can build certain tools that can facilitate traffic from new sources like ChatGPT and Perplexity.
How to optimize the content for AI?
Incorporating longer image descriptions, FAQs, and schema tags so that LLMs can read content better. Also, referencing and changing the way citations are done matters. An intent-based approach is needed.
AI crawlers do not read Dynamic content, is that true?
When you enable the search function in Open AI, the LLM launches an assistant that shoots the query to search engine partner providers (usually Bing). The system parses the snippets that the search engine provider knows about. If that snippet ranks, they will try to read the document that generated that snippet. It’s unclear if Open AI can index React client-side applications, but you can check the Open AI website documentation about Open AI crawlers or simulate an Open AI crawler to check if it can index the content. When it comes to the web part of LLMs, they’re still relying on more or less traditional search indexation techniques.
How should we approach SEO when some users rely on LLMs while others still use Google Search?
Think about it as the early stage of the internet when people had different browsers. Now, people are using different LLM models or different LLM assistants. We still have a traditional way of people searching for information, and we have the new emerging pack where people are talking to assistants. Still, we are solving mostly the same problem. For traditional SEO, don’t disregard it. Still follow Google Webmaster guidelines. But keep an eye on what’s going on with LLMs. It’s an emerging channel that can give traffic and leads in parallel with SEO.
How do you determine search intent versus keyword research?
In the traditional SEO world, user intent is a combination of semantical parameters matched with a database. It’s keyword-based. In the LLM world, you optimize for vectors. Every search query/conversation is tokenized, and LLMs look at the closest things that are relevant. When working with determining what to focus on, keep in mind that LLMs don’t operate on keywords. To optimize for LLMs, give more context and use language-based tools and a better structure.
It’s always said that we should create content for humans and not search engines. Is there a shift to Middle Ground as ultimately the end-user is still human but AI is now the processor of the information?
LLMs give favor to well-structured content that gives definite answers to users’ questions. There’s some sort of middle ground, but you can structure content in a better way and back it up with better references, which in return will make it better for humans too. Those optimizations usually go hand in hand.
With LLM answers vary between users. What should be our content approach in SaaS?
LLM-based platforms are unlikely to generate the same answers for the same people because they are predictive machines. The content approach should be to give explicit, specific answers to a range of queries. Think about using more long-tail keywords when thinking about content, and try to see which answers you can give to people, provided they will be reading it a little bit differently.
Is there a contradiction between LLM and search recommendations regarding keyword stuffing?
The scientific research shows that keyword stuffing, which could be good for SEO, doesn’t help LLMs. They lose context. When you stuff your blogs or documents with keywords, they rather pick something else that will be more understandable for them.
How do we educate ChatGPT about our business?
The more information you put into the systems, the more they know about you. Building custom GPTs on a knowledge base of your business is recommended.
Recommendations for learning more about marketing in times of LLM?
If you already have a good marketing background, educating yourself about LLMs would be the best next step. Learning about how LLMs operate is recommended.
