The word of the year in B2B Marketing and Sales: Context!

Watch what's actually breaking across B2B right now and you'll see the same root cause everywhere.

• You optimise your content to show up in AI answers — but buyers aren't asking the questions you think they are, and the people asking them aren't who you think they are either. Context problem.
• Your CFO asks an AI tool to evaluate your solution and gets a different answer than your internal champion did. Same vendor. Same product. Different context, different verdict. Context problem.
• You bolt AI onto your rules based automation platform and the output is still generic. Because the engine underneath still doesn't know who it's talking to. Context problem.

We spent the last decade building infrastructure for volume: More leads. More touches. More sequences. More "personalization" that was never actually personal. More (lost) opportunities.

And now we've added AI on top of that same foundation. As a result we most likely soon will find out none of it is really moving the needle the way we expected it to.

AI doesn't fix a context problem. It amplifies it. Give it shallow input and it produces shallow output at 10x the speed.

Argument 1: AEO — you're optimizing for the wrong questions

Take AEO — Answer Engine Optimization — the thing everyone says is replacing SEO. The simplified version goes like this: make sure your content shows up when someone asks an LLM a question. Sounds reasonable.

But most marketers are already getting the first part wrong. They're optimizing for the wrong questions. Nobody in an active enterprise buying process opens ChatGPT and asks: "What are the five best enterprise ERPs?"

That's not how buying actually works. Early in the journey, the phase where you can actually shape perception, buyers are asking very different questions. Operational questions. Problem-framing questions. Questions they haven't yet mapped to a vendor category.

That's the phase that matters. That's where you need to show up.

And those questions are different for every member of the buying committee. The CFO is asking something different from the IT lead. The procurement manager is asking something different from the end-user champion. If you don't know what each of them is actually asking, you're not doing AEO. You're doing content production and calling it LLM strategy but it's mostly out of context.

Argument 2: The LLM you're optimizing for doesn't exist

Our friend Kerry Cunningham, Head of Research at 6sense and one of the sharpest minds on B2B buying behavior research just published something that should stop every marketer in their tracks.

His team ran a research project asking LLMs to rate B2B vendor websites. Same instructions. Same websites. Multiple rounds.

After nine rounds of scoring, they still couldn't get consistent results across LLM instances. His conclusion: LLM outputs are not stable properties of your content. They're properties of your content in context.

The same question, asked by different users, gets different answers — because each user's LLM instance has been shaped by their history with it. Their usage patterns. Their professional framing. The way they habitually ask questions.

Kerry's example is sharp: a CFO who uses ChatGPT to stress-test financial models gets a more skeptical read of your ROI claims than your primary buyer persona does. Her instance has been conditioned to push back. And when she uploads your buyer's summary table and asks for a critique — she may quietly undermine the consensus your champion was trying to build.

You can't content-optimize your way around that.

This is what Kerry calls the shift from a content problem to a context problem. The implication: we may need LLM personas the way we built buyer personas — models of how different users' LLM instances are likely to behave based on who they are and how they use the tool.

Most organizations haven't started asking that question. Have you?

Argument 3: Your infrastructure can't use context even when you have it

Here's the argument that IMO doesn't get made enough. Even if you solve the context problem on the data side, even if you know who's in the buying committee, where they are in the journey, what they're asking, and what their LLM instance is likely to do with your content, most B2B infrastructure can't act on any of it.

Because it's rules-based.

Marketing automation, CRM workflows, lead scoring, nurture sequences — almost all of it runs on if/then logic. If a lead downloads X and opens Y, update field Z. Trigger campaign B. Move to stage 3.

Jon Miller, who literally helped invent the modern marketing automation category, calls these systems "rules engines wearing a user interface." That's not a criticism of execution. It's a statement about architecture. They were designed for a world where you could map the buyer journey in advance, pre-program every branch, and trust that reality would follow your flowchart.

It never did. But the system was fast enough and cheap enough that we made it work a little bit..... At least we were bringing some structure in our prospect engagements compared to what we were doing before.

The problem today is that buyers evolved even more and context-driven buying doesn't fit in a flowchart. A buying committee of seven people, each at a different point in their own journey, each asking different questions of their own LLM instance, each needing something different from your brand — that's not a workflow. That's a dynamic system. And dynamic systems need reasoning, not rules.

And here's what matters right now: some vendors are responding to this problem by wrapping AI around the outside of that same rules-based engine. An AI assistant on top. AI-suggested workflows. AI-generated copy fed into the same lead-centric, if/then infrastructure underneath.

That's not a solution. That's a paint job on a building that needs to be torn down.

The architecture is still wrong. The system is still organised around a single lead, a linear journey, and a set of rules someone had to think up in advance. You can add as much AI as you like on top of that, it will still execute the wrong model faster.

What Jon argues, and what I think is correct, is that the real shift is to Context AI: systems where you give the AI a goal, feed it deep business context, and let it reason its way to the right action for this person, at this moment. Not the action you pre-programmed. Not a rule. A decision.

This is what some of us are starting to call context engineering — formalizing the institutional knowledge, buyer signals, and operational data the AI needs to make intelligent decisions. Not prompts. Not playbooks. Architecture.

That's a fundamentally different operating model. And bolting AI onto a Marketo workflow isn't it.

Keep experimenting but be realistic about the outcome...

Here's what I keep seeing across the market:

Companies are investing heavily in AI-powered tools without first asking the harder question: what context do we actually have, and is it good enough to work with?

In B2B, context means: who is in the buying group, where are they in the journey, what do they already know, what are they worried about, what is each of them actually asking — and does your infrastructure have any ability to act on the answer?

Almost no company has that cleanly. Most have fragments. Disconnected signals sitting in five different platforms, owned by three different teams, none of them talking to each other, and all of it flowing into systems that can only execute what someone already thought to program via a lineair flow of pre designed treatments.

So the AI takes the fragments and does its best and the output lands like a stranger who knows your name but nothing else about you.

Experiment. You should. Everyone should.

But...the companies that will pull ahead are not the ones running the most engagement pilots. They're the ones that don't let experimentation become a substitute for thinking.

Because the context challenge doesn't go away while you're testing prompts and skills. If anything, every experiment that ignores the context challenge pushes the problem further out in front of you — and makes it more expensive to solve later.

The winners are the ones asking harder questions right now:

• What does our context architecture actually look like, and is it fit for AI reasoning, not just rule execution?
• Are we thinking seriously about MCP infrastructure, the layer that connects context to action across systems?
• Do we really understand our buyer's buying processes?
• Are we thinking about the impact of the tokenomics, what it costs to feed rich context into models at scale, and whether that math works for our business?

Some Caution

And, perhaps most importantly, are we thinking about reliance? Because Anthropic's Fable situation last week should have been a wake-up call for every (B2B) team that thinks that a public cloud model should be central to their GTM operations. A frontier model, serving hundreds of millions of users, taken offline within hours of launch. A Chinese competitor cited the shutdown the next day as proof that US AI models can't be relied upon — and watched its stock surge 33%. Note: I'm not promoting Chinese AI here. Not.At.All. I'm just pointing to the reliance impact. Because..:

Do we really need public cloud-based frontier models for all of this? And do we for every use case? Or do we need to think harder about where sovereign, local, or private model/open source infrastructure belongs in a serious B2B context stack? What about model routing? When do we rely on frontier AI and when on our own specialized models, building OUR buyer context, leveraging our first and 3rd party contextual data? This is not just an economics question, it's also about training your model correctly.

These aren't questions most business user teams are asking. They're just running pilots. Good. But the depth of context you accumulate, centralize and build in your relationship with the buyer — that is the new competitive moat.

Being in the marketing department using Claude in a vacuum or thinking that connecting Claude to Marketo's MCP layer ingesting dirty data out of context, is a viable solution while others in the enterprise are thinking about managing (often restricting) AI data and infrastructure far away from the reality of the business processes that's creating a new divide. A functional alignment problem all over again. I think we need to talk to IT and IT needs to talk to the GTM teams urgently.

I'd welcome your thoughts and ideas.

#B2B #GTM #AEO #BuyingGroups #ContextAI #MarketingAutomation #MarketingStrategy