In March 2026, Anthropic published research introducing a new metric called “observed exposure” — a measure of the gap between what AI can theoretically do and what people actually use it for.
For computer and math workers — the most technically literate, AI-savvy profession on the planet — large language models can theoretically handle 94% of their tasks. But actual observed usage covers just 33%.
The most AI-native profession in the world has closed barely a third of the gap.
If that’s the benchmark, where does marketing stand?
Marketing teams are at roughly 5%
Let’s be honest about what “using AI” looks like in most marketing teams today. Someone generates a blog post draft with ChatGPT. Someone else creates social images with Midjourney. A few people use AI to summarise meeting notes or rewrite email subject lines.
This is single-model, single-task, manually-triggered AI usage. It’s useful, but it’s the equivalent of using a smartphone only as a calculator.
There’s nothing agentic. Nothing autonomous. Nothing that compounds.
Most marketing teams are using roughly 5% of what AI can already do today. Not in five years. Today.
What 50% looks like
To understand the gap, you need to see the other side.
Imagine continuously running AI agents that are analysing your acquisition data, generating new hypotheses, designing test variations, shipping those tests, and reporting back with results — completely closed loop.
Now imagine that running across every channel simultaneously. SEO. Email. Paid. Organic social. Referrals. GEO. Each channel with its own agent, its own experimentation loop, its own compounding learning.
The underlying technology exists today. The teams that are building toward this are already pulling ahead.
The gap between where most marketing teams are (single-model content generation) and where they could be (autonomous, closed-loop experimentation) is growing every day.
The leadership spectrum
The reason most teams are stuck at 5% has nothing to do with the technology. It’s a leadership problem.
Across the companies I work with, I see a spectrum:
- AI is banned. No AI tools allowed. Full stop.
- AI is allowed, with constraints. Teams can use a single approved tool, but with no MCP connectivity, no CLI access, and no ability to connect internal tools or client data.
- AI is mandated. Usage is a core KPI. There’s a top-down mandate to integrate AI into every workflow.
Most companies sit in the middle — AI is allowed, but constrained. And this is the most dangerous position, because it creates the illusion of progress. The team “uses AI” but operates under restrictions that prevent any real capability from being built.
Meanwhile, companies at the far end of the spectrum are moving fast. When Shopify CEO Tobi Lutke told his team that reflexive AI usage is now a baseline expectation — and that teams must demonstrate why AI cannot do the job before requesting additional headcount — he wasn’t making a suggestion. He was redrawing the competitive landscape.
Using AI well is a skill that needs to be carefully learned by… using it a lot.
Tobi Lutke, CEO at Shopify
The gap between constrained-middle companies and mandate-driven companies isn’t narrowing. It’s widening fast.
Random acts of AI
If you’ve followed our writing, you’ll be familiar with the concept of random acts of marketing — the spray-and-pray approach where teams run disconnected activities with no shared focus, no process, and no way to measure what is working.
AI adoption in most marketing teams is following the exact same pattern.
One person uses ChatGPT for headlines. Another uses Claude for summarising customer calls. Someone in demand gen is experimenting with AI-generated ads. The content team has its own workflow. None of it is connected. None of it compounds. There’s no system.
This is random acts of AI. And it’s the new disease killing marketing team performance.
The symptoms look familiar: lots of activity, very little measurable impact. Individual productivity gains that never translate into team-level or business-level outcomes. The feeling that everyone is busy with AI but nothing has fundamentally changed.
The cure: outcomes over outputs
The cure for random acts of AI is the same as the cure for random acts of marketing — focus on impact over activity.
It doesn’t matter how many AI tools your team has access to, how many prompts they’ve saved, or how many hours they’ve “saved” with AI-generated content. What matters is whether your core marketing metrics are improving: pipeline, conversion rates, revenue.
Outcomes over outputs. Always.
This means:
- Start with the metric, not the tool. Don’t ask “how can we use AI?” Ask “what’s the biggest constraint on our pipeline right now, and can AI help us run more experiments against it?”
- Build a system, not a collection of shortcuts. AI should slot into your growth marketing process, not exist alongside it. Every AI-powered workflow should connect to a hypothesis, a test, and a measurable outcome.
- Think agentic, not assistive. The real unlock isn’t AI that helps you write faster. It’s AI that can autonomously run experimentation loops — analyse data, generate hypotheses, design tests, and report results without manual intervention at every step.
- Make it a leadership priority. If AI adoption is left to individual team members experimenting in their spare time, you will get random acts of AI. It needs to be a strategic priority with clear goals, resources, and accountability.
The stakes: flat today, declining tomorrow
Here’s what getting left behind looks like right now: stagnation. Everything is flat. Leads, trial signups, conversions, meetings booked — all flat. It’s getting harder to move core marketing business metrics, and the usual playbooks are delivering diminishing returns.
The Anthropic research confirms what many of us are feeling — we’re at the very beginning of this shift. The study found no systematic increase in unemployment yet, but BLS growth projections for AI-exposed occupations are already weakening. The leading indicator isn’t job losses. It’s a growing performance gap between teams that have built real AI capability and those running random acts of AI.
AI is far from reaching its theoretical capability: actual coverage remains a fraction of what is feasible.
Anthropic, Labor Market Impacts of AI
My prediction: marketing teams that aren’t AI-first will move from flat to decline over the next 12 months. Not because AI replaces them, but because AI-first competitors will simply out-experiment them at every turn.
The gap between the AI-haves and the AI-have-nots is the defining competitive dynamic of the next few years. And for marketing teams specifically, the clock is ticking.
Your team is probably “using AI.” The question is whether you’re building a system — or committing random acts of AI.