You Gave AI Your Marketing. You Forgot to Give It Your Business.

Here is a number that should bother you: 98% of marketing teams now use AI. And here is the number that should bother you more: only 10% report transformational results.

That is an 88-point gap. Nearly everyone adopted the technology. Almost nobody got what they expected from it. The natural instinct is to blame the tool. Maybe the AI isn't good enough yet. Maybe you picked the wrong platform. Maybe the next model update will fix it.

It won't. Because the problem was never the AI. The problem is what you plugged it into.

The Horse, the Cart, and the Missing Piece Between Them

In February 2026, researcher Ethan Mollick introduced a framework that quietly explains the entire gap between AI promise and AI performance. He broke "using AI" into three layers:

• The Model is the intelligence. Claude, GPT, Gemini. The brain.
• The App is the interface. The chatbot, the website, the phone app. How you talk to the brain.
• The Harness is the system that connects the brain to real work. The tools, data, context, and infrastructure that let the AI actually do things rather than just say things.

Here is the insight most businesses missed: the big three AI models are now remarkably close in overall capability. Mollick himself says as much. The model differences are "small enough that the app and harness matter more than the model." You are not losing to competitors because they picked a better AI. You are losing because they built a better harness.

And your marketing infrastructure, whether you realised it or not, IS your harness.

Your Marketing Already Has a Harness. You Just Starved It.

Think about what your AI actually has access to when you ask it to write an ad, plan a campaign, or draft an email.

For most businesses, the answer is: a prompt and maybe a URL. You type "write me a Google Ads headline for my plumbing business in Adelaide" and the AI generates something competent, generic, and identical to what it would produce for every other plumber in Adelaide.

Now think about what the AI could have access to:

This is what Mollick means by harness. The model is the same. The harness changes everything.

Context Is Multiplicative, Not Additive

Rory Sutherland, the behavioural economist and Vice Chairman of Ogilvy, has spent decades arguing that context doesn't just change perception. It multiplies it. His favourite example: the taste of food is affected by restaurant decor, music, glassware, the waiter's confidence. "If one critical element is zero, it reduces the entire experience to zero regardless of food quality."

The same principle applies to your AI's marketing output. The model is the chef. The harness is the restaurant. You can have the best chef in the world, but if you seat them in a bare room with plastic cutlery and fluorescent lights, the food tastes worse. Not metaphorically. Measurably.

When your AI writes marketing with zero customer context, zero performance data, and zero competitive intelligence, the output isn't just missing those inputs. It actively produces the opposite of what you need. It generates the average. The expected. The safely mediocre. Every piece of context you withhold is a multiplier set to zero.

Sutherland puts it bluntly: "It doesn't pay to be logical if everyone else is being logical." When 98% of marketing teams use AI to produce marketing, and 88% of them are feeding it the same generic inputs, the logical output is the same generic marketing. The only escape is feeding AI something nobody else has.

Your business data, your customer language, your competitive position, your conversion history. That is the non-replicable context. That is what turns the multiplier from zero to ten.

The Distinctiveness Problem: When Everyone's AI Sounds the Same

Byron Sharp's research at the Ehrenberg-Bass Institute across 130+ brands demonstrates a finding that becomes terrifyingly relevant in the AI era: brands grow by being easy to think of, not by being better. His term for this is mental availability, the propensity for a brand to be noticed or come to mind in buying situations.

Mental availability requires distinctiveness. Not differentiation (being objectively better), but distinctiveness (being recognisably different). Your colours, your language, your style, your specific way of framing the problem. These are what Sharp's colleague Jenni Romaniuk calls distinctive brand assets.

Here is where AI creates a paradox for businesses stuck at the Early and Emerging stages of maturity. When you use AI as a generic content machine, typing prompts into a chatbot and publishing what comes back, you are systematically destroying your brand's distinctiveness. The AI converges toward the average of its training data. Your ads sound like everyone's ads. Your emails read like everyone's emails. Your landing pages blend into the wallpaper of the internet.

We've written before about why AI-generated marketing often fails to stick. The mechanism is exactly this: zero-context AI produces zero-distinction marketing, which produces zero mental availability.

But the reverse is also true. When you feed AI your distinctive inputs, your customer language, your brand voice patterns, your specific case study data, the output inherits that distinctiveness. AI doesn't flatten your marketing. A starved AI flattens your marketing. A well-fed AI amplifies what makes you different.

The practical test: could your competitor copy-paste the same prompt and get the same ad? If yes, your harness is broken.

The Compound Effect: Why the Gap Widens

The software engineering community discovered something in 2026 that marketers haven't caught up to yet. Kieran Crowley at Every documented a framework called Compound Engineering where each unit of work makes subsequent work easier. His team cut feature development time from weeks to 1-3 days by investing 50% of their time in codifying patterns, building review systems, and documenting solutions rather than just writing more code.

The marketing parallel is direct. Every campaign you run generates data. Every customer interaction reveals language patterns. Every conversion tells you what actually works. The question is whether any of that knowledge is captured and fed back into your AI's harness, or whether it disappears into a dashboard nobody checks.

Simon Willison, one of the most prolific AI practitioners working today, observed that "managing agents is really a management problem." The skills that make AI effective, clear briefs, measurable objectives, quality control, feedback loops, are the same skills that make any team effective. They just compound faster with AI because the cycle time is minutes instead of weeks.

The businesses at the Transformational stage didn't arrive there overnight. They built the harness one piece at a time. First, they connected conversion data. Then customer feedback. Then competitive intelligence. Then brand guidelines. Each addition made the AI's output measurably better, which generated better data, which further improved the harness.

This is the real danger for businesses stuck at Stage 1. The gap isn't static. It compounds. The businesses ahead of you aren't just a little ahead. They're pulling away at an accelerating rate because their AI learns from better data while yours learns from nothing. We explored this compounding dynamic in how AI usage went up 10x while staying at the same level. The pattern hasn't changed. It has accelerated.

What This Means for Your Business

You don't need a better AI model. You need to give the one you have something to work with.

• Week 1: Export your top-performing search terms, best customer reviews, and highest-converting service pages. Create a reference document your AI can use.
• Month 1: Connect your ad performance data. Show AI what actually worked, not just what you think should work. Your cheapest leads might be costing you the most, and only real data reveals which.
• Month 2: Add customer voice data. Call recordings, review language, enquiry patterns. This is where your best ad copy lives, and it is the single most underused input in AI-driven marketing.
• Month 3: Build feedback loops. Every campaign result should improve the next brief. Every customer response should refine the voice. The compound effect starts here.

Further Reading

• A Guide to Which AI to Use in the Agentic Era by Ethan Mollick. The original model/app/harness framework that explains why 98% adoption produces 10% transformation.
• Compound Engineering: The Definitive Guide by Kieran Crowley. How each unit of AI-assisted work can make the next one better, applied to software engineering but directly transferable to marketing operations.
• Context Engineering Is the Real AI Advantage in Marketing from MarTech. Why "context engineering" is replacing "prompt engineering" as the skill that determines AI marketing performance.
• Alchemy: The Surprising Power of Ideas That Don't Make Sense by Rory Sutherland. The foundational argument for why context multiplies value, and why the logical approach guarantees the same results as everyone else.
• Building Distinctive Brand Assets by Jenni Romaniuk. The Ehrenberg-Bass research on what makes brands memorable, and why consistency of distinctive assets matters more than quality of individual executions.