Stop treating AI like a search engine if you want to win marketshare

Generative AI is structurally changing the B2B tech buying journey. From discovery to recommendation, we are now turning to our favourite AI assistant to help us make the right choice - or at least reduce the risk of making the wrong one. But winning market share in this new environment means ditching the old playbook. Instead of fighting for bottom-of-funnel citations, brands must shape the AI's understanding of their category from the ground up. 

Click-through rates have collapsed by up to 61%, according to the latest study by marketing agency Seer Interactive. And referral traffic from the machines is steadily increasing, according to Previsible.

Marketers and comms professionals are now being tasked with winning at AI. “How do we make sure we’re the ones being cited” is a question being demanded by CEOs and boards alike. And they aren’t the only ones hearing this call to arms. Every other week there is a new influencer or vendor touting the latest technical trick - like hidden text or complex schema - to influence the engines.

But the reality is, there is no silver bullet. No magic trick. And, even if there is, the algorithm is changing so fast that you leave yourself vulnerable to emerging cliff-edges.

So, what does work? I sat down with Tom Rudnai, founder and CEO of Demand Genius in our Back2Brave podcast to ask that exact question. He’s been building a solution that looks at AI visibility from a much more holistic perspective and has a unique understanding of how these large language models actually work.

Why AI search breaks the traditional B2B tech buyer journey

Traditional SEO has been a mainstay in marketing tactical executions almost since the internet began. The aim: to make page one of Google so that when a buyer comes to market, they discover and hopefully buy from your brand. The higher your blue link in the rankings, the more traffic to your website.

But AI models have flipped this dynamic completely. They don’t care where your blue link ranks on Google. Instead they synthesise answers from their training data and any relevant information available to them, crawling hundreds of thousands of blue links and web pages in an instant. Having to open 10 blue links to research a category (because let’s face it, who really opened more than that), is no longer a requirement on the hunt for a vendor or solution to a problem. Just open up ChatGPT or Claude (which is filled with your personal context), ask a few questions without having to leave the platform and you have a fairly good idea of what path you’re going to take.

According to Tom, and his recent Demand Genius Dark AI report into algorithmic retrieval, the research phase has probably seen the biggest impact (and drop in traffic to your website). The report states AI retrieves external citations 16% of the time in total, meaning it answers from memory 84% of the time overall. During the awareness and consideration phases, retrieval is 0%, meaning it answers from memory 100% of the time.

“What we were doing was analysing how AI responds at awareness, consideration, and conversion stage of the funnel to try and understand what behaviors we see at each stage. At the start it's very exploratory. It then becomes very comparative and then very directive. So it goes from exploration mode into helping you make a decision as you move down the funnel.” 

“It's what we call intent matching. If they're trying to learn, it helps them learn and it explores. If they're trying to make a decision, it helps them get towards a decision. And it changes the way it behaves based on the intent. So citation and retrieval rate at a basic level is 0% in awareness, 0% in consideration and 48% at conversion. If you're doing GEO like SEO, what you're doing is you're optimising for those 48% of queries where AI invokes retrieval and then goes to a behavior that could be considered equivalent to search.”

What does this mean for B2B tech brands? It means that if you are only optimising to be cited, you're fighting for Share of LLM right at the bottom of the funnel. And, to be honest, your chances of being cited are slim because the LLM will lean more heavily towards the brands that have educated it on the category, the pain points and the solution. 

Tom uses the analogy of an RFP to make this point.

“I describe myself as a recovering sales rep, and there was this cliche that if you get sent an RFP you’ve already lost it because someone else was in there helping them to write and define it. I could look at an RFP and tell you who was going to win it because I could tell who had helped to write it.

“And that’s the real challenge and opportunity of GEO, because if you can produce really high-quality research that influences our collective understanding of a category, then you can actually expand your TAM and more in-market buyers get pushed in for your product. Because if you can get a piece of content into the ingrained understanding a model has of how your category should function and what’s important, then it’s going to be pushed back into every single conversation.”

The key takeaway: shape the category, not the end recommendation. This early phase of the buyer journey establishes the buying criteria. If you're shaping an AI’s internal understanding of the category early, the chances of you making the final shortlist when citations actually appear is exponentially greater.

Why AI search engines only reward information gain

So, I hear you ask, how do I influence a model? The simple answer is giving it something new that it doesn’t already know.

Generative engines like ChatGPT actively filter out generic, homogenised content. These models ignore anything that adds nothing new to their training data, choosing instead to reward distinct empirical and interpretive information gain.

What is information gain? Tom defines it as whether a piece of content adds net new understanding to a problem or space. Basically, if an LLM could generate the same piece from existing sources, you’re not contributing information gain.

And it comes in four levels:

• Level 0: No information gain
• Level 1: Interpretive gain - takes existing ideas and reframes them, adds opinion or industry insight, or is a contrarian take backed by evidence.
• Level 2: Empirical gain - original data or research that advances its understanding.
• Level 3: Conceptual gain - introduces a new mental model, framework or taxonomy.

Most B2B tech content today lives at Level 0. No matter how well written that blog piece might be, if it has been explained a thousand times before, it’s not adding anything new.

What does this mean for B2B tech brands? Well, if your content strategy relies on churning out high volumes of generic - dare I say it - AI generated content, you are wasting budget. Not only because they add zero value to a model but also to the reader. Want to check? Just have a look at the traffic numbers to your blog pages and you’ll soon find out which ones might have been better off being left unwritten.

Today, it’s about quality over quantity and there are three things you should be looking to achieve when writing your content:

1. Extract unique insights: look at your own operations to create empirical gain. Gong Labs turns internal sales call data into unique market reports. Do the same with your platform data. If you can’t easily extract the research, develop and commission some unique research into the areas you want to own.
2. Have a brave and distinct opinion: Create interpretive gain by taking a brave stance, a contrarian take or providing unique industry insight. If five competitors offer identical software solutions, the brand with the brave take will be cited by LLMs to provide necessary contrast in its synthesised response.
3. Maintenance matters: ditch the volume game and shift your focus to maintenance. Your content estate matters, so ensuring it’s clear, consistent and compelling is the top priority. You then supplement this with fewer, high-impact pieces, backed by data, which challenge the status quo.

Content debt confuses the LLM and consumes more compute

Content debt is the years of contradictory blogs, outdated positioning and unmaintained content that sits in your content estate. Because an AI model evaluates your entire content footprint, having contradictory, conflicting or outdated information creatives a massive liability for B2B tech brands. 

This not only creates significant issues in LLMs understanding your brand today, it could be detrimental when compute is no longer subsidised. 

Let me explain. 

When an LLM is looking to understand a brand or topic, it is looking for consensus. If you’ve pivoted your strategic positioning every six months or have unchanged content from back when you were a scale up, the AI receives conflicting signals. That’s not to say it can’t work out the history of the brand, but you are making it harder to ascertain the truth about your organisation. The impact - your messaging gets diluted, there is a greater change of hallucinations.

What’s worse is that when platforms like Claude and Perplexity stop being subsidised by investors, the compute required to understand your brand might see LLMs prioritise easy answers. Now this is a hypothesis and I’m sure LLMs will try to avoid bias. But if we look at it logically - the more processing and compute required to understand your brand will cost the models more money to process. When they transition to become profit hungry, the models could deprioritise anything that is hard work in favour of content that makes it easy for them - i.e. requires less compute to process.

While this might be an educated guess, the sentiment is still important. The easier it is for LLMs to understand your brand, the more likely they are to inject you into relevant conversations in the right way. 

The key takeaway: The old model of producing high volume of net new content for SEO purposes is dead. You still need new….just not as much. And it needs to be higher quality and most importantly add information gain. 

Content budgets should move from having a heavy focus on production to more of a 50:50 split. Half your focus should be on great new content. The other half should be spent on maintenance to ensure you are presenting a single, unified narrative across all your channels.

As we advise our clients at Babel, clarity and consistency are your unfair advantage when it comes to GEO.

Why B2B tech brands must return to marketing fundamentals for GEO success

There are no quick hacks to GEO. No backlink strategy. No keyword stuffing. No technical schema workarounds. While some might work for a while, the LLMs are constantly shifting their algorithms which could make all that work redundant in a heart beat and you’re back to square one. 

What matters is strong positioning, clear narratives, and distinctive differentiation - the foundations of core marketing principles. B2B tech brands that invest in building a strong brand, with brave perspectives backed by original data will outlast any algorithmic update.

So, start there. Do you have a solid brand positioning? Is it easy to understand, differentiated and distinctive versus your competitors? Build out what I call your desired knowledge profile - essentially what would you want an LLM to say about you, in which situations would you like them to say it and what category entry points do you need to influence and own.

Once you have that, audit your current content estate for conflicting messaging. Now, content estates can be massive, so I’m not suggesting you overhaul it in a week. But rather, prioritise your content against the below pillars:

Protect

High-value content that should remain.

Optimise

Content with value that requires improvement.

Refresh

Content requiring updated information.

Consolidate

Multiple assets covering similar areas.

Retire

Content that creates little value or potential confusion.

Create

Strategically important gaps requiring new content.

Small plug here for Tom’s platform - this is what it was designed to help with and its Content Intelligence functionality is second to none (in my opinion). It’s what we use when managing a client’s content estate.

Now that you have a content plan in place, you can turn your attention back to net new content. Make sure you have a good mix of top of funnel (category influence), middle of funnel (problem understanding) and bottom of funnel (cite me for this) content. Vet every piece of content for originality, uniqueness, differentiation and, most importantly, against the four levels of information gain.

If you are ready to stop playing it safe and want to ensure your brand is positioned to win in an AI-mediated market, get in touch with the Babel team to discuss our dedicated GEO services.

Want to watch the episode → YouTube

Want to listen to the episode → Spotify | Apple Podcasts | Babel Website

‍