The Intention Architect: The Senior Marketing Skill That Replaces Attention Work
The role replacing attention work has no title yet. Senior marketers who name it first can own it.
Another mix of different articles and published papers that help marketers understand how marketing work is changing.
Last month I started reading a paper that I now think every senior marketer can use. It came from the Shorenstein Center at the Harvard Kennedy School, which is not where one would usually look for martech direction... The paper is called From Attention Merchants to Intention Architects. Its argument is structural and it brings an interesting perspective for marketers. Please, stay with me!
As always just reply to this email and let me know your thoughts, so we can keep this convo going.
The attention economy seems to be collapsing. Major publishers lost fifty percent of their traffic when Google added AI overviews. The $685 billion digital advertising industry faces existential pressure as AI assistants stop clicking on ads. Search engine optimization began being questioned the moment search engines stopped sending people to websites. This is happening in measurable, audited numbers across every channel the modern marketing playbook was built on.
What replaces it is what the Shorenstein authors call the intention economy. The shift, in their words: the question is no longer who controls what we read. The question is who controls what we think to ask.
For senior marketers, that sentence should land like a fire alarm (also, as a human and citizen this concerns me even more - but that is a topic for another article).
What intention work actually is
Attention work is interruption work. You buy media to put your brand in front of a customer who was looking for something else. You optimize for the click, the open, the impression. You measure success in how often you got in the way of someone else’s flow.
Intention work is anticipation work. You build the infrastructure that decides what surfaces when a customer asks an AI agent for help. You shape the structured data your category appears in. You design the consent and identity layer that determines whether AI can act on your customer’s behalf at all. You measure success in whether your brand is the answer to a question your customer will ask tomorrow.
This paper also confirms my point about how high quality data is the foundation for this intention work to happen. And governance is what creates processes that support the collection and activation of this high quality data in a sustainable and scalable fashion.
I believe that, in practice, the role that does this work has no title yet. The Shorenstein authors call it “intention architect” almost in passing. Boston Consulting group (BCG), in their Agentic Scenarios framework published two months ago, calls it discoverability work and desirability work and lists both as non-negotiable across every possible future of the marketing function. McKinsey calls it agentic readiness. None of them calls it a job or title.
That gap is the opportunity.
You are reading The Experimental Marketer Newsletter.
What you get: every week, one martech or AI concept broken down
with real examples, step-by-step walkthroughs, and the career case for why it matters.
Created by a Marketer, for Marketers.
The Ballantine’s anchor
Here are some examples on how this is impacting brand work in 2026.
Gokcen Karaca, head of digital and design at Pernod Ricard, learned that two-thirds of Gen Z and more than half of millennials had started using large language models to research products. He partnered with a firm called Jellyfish to analyze how leading AI models represented Pernod Ricard’s brands. The findings were not flattering. One popular AI model had miscategorized Ballantine’s Scotch — a mass-market, affordable offering — as a prestige product. Harvard Business Review reported the story this March.
Now, let’s think about what that means operationally for marketers. A customer asks an AI assistant for an affordable, sharable Scotch for a dinner party. The model does not surface Ballantine’s because it thinks Ballantine’s is prestige. A different customer asks for a special-occasion gift. The model surfaces Ballantine’s at a price point the brand itself never set. Both customers leave the conversation with the wrong information. Both purchase decisions are nudged in directions the brand cannot see, cannot correct, and cannot measure with existing analytics.
The brand was being indexed without anyone in the marketing department knowing what the index looked like.
This is what intention work can prevent. Intention architects are not smarter than attention marketers. They are looking at a different surface.
Three Moves Senior Marketers Can Make This Quarter
There is no point in writing a piece like this without naming the marketing work. Here is the sequence I would run inside a marketing organization that wants to take intention architecture seriously, in the order I would run it. Each move is something a senior-level marketer can scope, and complete inside one quarter without asking for a new budget.
Move 1: Audit how your customer data is structured for agent-readability
Most enterprise customer data was built for dashboards. Dashboards read aggregated, post-hoc, batched data. Agents read live, structured, queryable data. Those are different design jobs.
Gabrielle Boko, CMO at NetApp, named the four preconditions for AI-ready marketing data in a Forbes piece earlier this year. Governed (you know what data you have, where it lives, who can use it, and what it can be used to train). Intelligent (your storage layer auto-tiers so you are not paying egress on cold data). Resilient (if customer data is compromised, your AI capability does not go with it). Mobile (data can move between environments and across clouds without losing fidelity or governance in transit).
Run the audit in three steps inside one quarter.
Map your customer data against the four preconditions. Score each one on a five-point scale based on what your team can actually do today.
Pick one customer intent and trace the data flow an agent would need. Re-engagement of a lapsed buyer. Cross-sell into a new product line. Try to articulate, end to end, the data an AI agent would need to act on that intent with full consent and governance intact. The places where the flow breaks are the gaps.
Produce a one-page use case you can share with other teams involved. Name the two preconditions where the organization is weakest and how to rectify it.
The work is upstream of any AEO or LLM optimization investment.
Move 2: Run the Ballantine’s test against your top three brands
This is a diagnostic audit, not a measurement system. The output is qualitative. Patterns, anomalies, and the obvious gaps you can take to your leadership this quarter. The audit is not a number you can defend to finance. Move 3 builds the measurement system. Move 2 tells you whether you need one urgently.
A word on the limitation that matters here. Large language models are probabilistic. Running one query against one model produces one possible output, not THE output. Run the same query an hour later and the response may differ in position, descriptors, competitors named, or sentiment. The output also varies by whether the user is logged in, whether web search is enabled, what prior turns sit in the conversation, and which model version is live that week. Single-point readings are unreliable as evidence. The audit handles this by building in repetition and by reading variance itself as a finding.
The audit has six steps.
Construct five queries the way a real customer would type them. Natural language, the way someone who does not yet know your category would phrase the question. Build the list with your customer insights team, if possible. Avoid language that hints at your brand’s positioning — query construction is the biggest source of self-confirming bias in this work.
Standardize the test conditions. Run all queries from the same network, in fresh sessions, logged out, with no prior conversation context. Run each query in two conditions: web search disabled and web search enabled, when the model offers that toggle. Disabled tests what is in the model’s training. Enabled tests what is in the live retrieval layer. These produce different gap profiles and need different fixes.
Run each query three times per model. Five queries times three models times three runs times two conditions equals 90 data points. Larger than an afternoon audit. Still scopable inside a one-week sprint for one analyst. The repetition is what separates a useful diagnostic from a misleading snapshot.
Document what stays constant and what varies. For each query, note three things. Does your brand appear consistently across the three runs, or only sometimes. When it appears, are the descriptors consistent. Does the competitive set rotate, or are the same competitors named each run. Consistency is itself a finding. A brand that appears in two of three runs is in a more fragile position than a brand that appears in all three.
Document the three types of gap. Invisibility, when your brand does not appear in queries where it should. Brand strategist Jess Graham, formerly of Instagram, calls this “Agentic Invisibility” and traces it back to weak “Algorithmic Legibility,” which is the discipline of making a brand easy for AI systems to read. Misrepresentation, when your category, price, or positioning is wrong, like the Ballantine’s case. Competitive disadvantage, when a specific competitor is consistently described more favorably than yours. The repetition tells you whether each gap is structural (consistent across runs) or volatile (varies across runs). Both are problems. They have different fixes.
Build the escalation memo. Which gaps require content correction, owned by your team. Which require technical signal correction, owned by your SEO and AEO teams/partners. Which require structured data correction, owned by your IT counterparts. Flag volatility findings explicitly. “Brand appears in 1 of 3 runs for this query type” is operationally different from “brand never appears.” The first is a signal-instability problem. The second is a training or indexing problem.
A note on what this audit does not tell you. It does not produce a defensible cross-model ranking — user populations differ across ChatGPT, Gemini, and Claude, for example, so direct comparisons are apples to oranges. It does not produce a number for finance. It does not establish a trendline. Move 3 does those things. Move 2 produces the diagnostic that justifies the investment in Move 3.
The marketers who run this audit first can become the people the CMO calls when the question gets asked at the next board meeting, and that question is coming.
Move 3: Establish a baseline measurement of your brand’s presence in AI-mediated discovery
BCG’s analysis found that traditional search results and AI-generated answers overlap by only eight to twelve percent. Meaning: if you have only optimized for search, you can be invisible to AI-mediated discovery in eighty-eight to ninety-two percent of cases.
That is not a tuning problem. That is a baseline measurement that does not yet exist in most marketing organizations, and the team that builds the measurement controls the conversation about what to do next.
Five steps to set up the baseline inside one quarter.
Define your category in the language a customer would use. The phrase they would say to an AI assistant when they have a problem your product solves. Use customer-spoken language only. Internal product names distort the test.
Define your competitive set. Your top five named competitors plus two adjacent disruptors the agents might surface in your place.
Establish the monthly snapshot. Once a month, run the same fifteen queries across the same three AI assistants. Capture which brands appear, in what order, with what descriptors. Store the results in a versioned location your team can return to.
Build the comparison view. Month over month, which brands are gaining visibility, which are losing, where the gap is widening or narrowing, and which AI model is most generous to your category and which is least.
Document the methodology before the results. Recent academic research on AI visibility measurement has shown that single-point measurements are unreliable. Variance is high across models, across runs, and across time. The 2026 arxiv paper “Don’t Measure Once” argues visibility must be measured as a distribution, not as a number. Write your methodology down. When results contradict each other across months, your team can interpret instead of react.
Three signals worth tracking inside the baseline. Whether your brand appears in answer-engine outputs across your top customer-query patterns. Whether the appearances are accurate. Whether the trajectory is improving or deteriorating across monthly snapshots.
The Career Math
Every framework I have cited in this piece — Shorenstein, BCG, HBR, McKinsey — has named the same gap but none of them has named the person who closes it. That gap will not stay open.
The marketing professional I write for already has the skills for this work. The transition is not a reskilling project. It is a renaming project. The marketers who understand the work behind the title first can be better positioned to claim the budget and the authority for the role.
Attention work is eroding. The role replacing it does not yet have a name your CMO would recognize.
Give it one before someone else does.