PART THREE — RUNNING IT
Chapter 12 — Paying for It
This chapter covers budget: what the 60:40 brand-to-activation ratio actually says, the legitimate critique of the databank behind it, where money should go across the five levers, and who inside the organisation should own each one — because most of this work has no natural home in a marketing org chart.
From Becoming the Answer by Jeremy Osborn · 1,145 words
The argument you’ll hear is that AI search requires shifting budget from performance to brand. The evidence for that is better than most claims in this field, and the canonical citation people reach for is weaker than they think. Know both halves.
What the 60/40 rule actually says
The 60:40 brand-to-activation ratio comes from Les Binet and Peter Field’s work for the IPA, drawing on its effectiveness databank. Its authors framed it as a guideline for the average case, not a law.
The critique is legitimate and hasn’t been rebutted on the merits. Byron Sharp of the Ehrenberg-Bass Institute has called the rule “very misleading,” built on “a very weird data set” — the databank is composed of award submissions, which are self-selected, self-reported campaigns that entrants believed had worked. His argument is that the number exists largely because the industry wanted a number.
The honest middle came from James Hurman: the rule may be less scientific than Sharp would like, but few would argue with the observation that marketers tend to underinvest in brand.
And it varies by context — its own authors say so. The optimal brand share runs from roughly 20 to 80 percent depending on how the category buys. For B2B specifically, the analysis of B2B cases in the same databank recommends closer to 50/50, with the authors describing their own findings as tentative and their samples as small.
So the defensible claim is directional, not arithmetic. Most organizations underinvest in brand. The right ratio for your category is not 60/40 because a book said so.
Which way budgets are actually moving
Against you, as it happens, which makes this argument contrarian rather than fashionable.
NIQ’s 2026 CMO survey of over 250 senior decision-makers found only 55 percent allocating 60 percent or more to long-term brand building, down from 59 percent the year before — and only 69 percent saying their C-suite believes in long-term brand value, down from 80 percent. Seventy-four percent report heightened ROI scrutiny.
Gartner’s survey of 401 CMOs found marketing budgets flat at 7.8 percent of revenue, 15 percent of marketing budget allocated to AI generally, 70 percent saying AI leadership is critical, and only 30 percent reporting mature readiness. No discrete line item for any of this.
So you’re making a case for less measurable investment at exactly the moment accountability pressure is tightening. Pretending otherwise won’t help you win the argument.
The argument that actually works
Don’t lead with 60/40. Lead with the mechanism.
One: most questions never trigger a search. Between two-thirds and four-fifths of prompts are answered from what the model already holds. For those, brand prevalence is the visibility mechanism. There’s no content tactic that reaches them. This is a brand argument grounded in architecture rather than in award-show data, and it’s the strongest thing you have.
Two: you can’t buy your way in. Using search as the reference, paid placement and organic recommendation stay structurally separate, because paid bias in organic results destroys the trust that makes the market work — especially in a market where users have demonstrated they’ll switch platforms readily. A platform whose share can fall twenty points in a year can’t afford to sell its recommendations.
State the counterweight rather than hiding it: ads are already appearing inside answers. OpenAI began testing them in early 2026, and by mid-year roughly a quarter of ChatGPT responses contained one, labeled and separated. The organic and paid layers are separating exactly as they did in search. What remains unlikely is paid influence over the organic recommendation.
Three: the correlational evidence points at brand signals, not content signals. Unlinked mentions track AI visibility around 0.66; backlinks around 0.22; content volume around 0.18. Correlation isn’t causation and the confound is real. But if you’re allocating under uncertainty, allocate toward the stronger relationship.
Four, and this is the one that closes. You aren’t asking for new budget. You’re asking to redirect existing spend toward better-evidenced mechanisms. Some content production money moves to original research, which serves both information gain and earned media. Some link-building money moves to digital PR aimed at the outlets that actually get cited, which produces mentions rather than links. Some technical budget moves to server-side rendering and crawl cleanup, which costs less than what it replaces. Entity and feed work is small, largely one-time, and unowned today.
Where the money goes
A defensible allocation for a mid-sized program:
| Area | Share | Why |
|---|---|---|
| Core search fundamentals | ~40% | Getting retrieved dominates everything after |
| Digital PR and earned media | ~25% | Strongest correlate; corroboration is a documented input |
| Measurement and reporting | ~20% | The sampling requirements are real and unmet |
| Enablement and training | ~10% | Most of your team’s mental model is three years old |
| Experimentation | ~5% | YouTube, agent-readiness, whatever comes next |
The line that will get challenged is the 20 percent on measurement. Defend it with Chapter 11: without adequate sampling you can’t tell whether any of the other 80 percent worked, and you’ll spend the difference arguing about noise.
Who owns this
The most common organizational failure here is assigning all of it to whoever owns SEO, when roughly half the work isn’t SEO work.
| Lever | Natural owner | Usual failure |
|---|---|---|
| Entity | Brand and corporate comms | Given to SEO, who can’t change LinkedIn or press boilerplate |
| Reachability | Engineering, with SEO | Given to SEO, who can’t change the rendering architecture |
| Corroboration | PR, analyst relations, community | Given to content, who can’t place earned media |
| Specificity | Content, briefed by sales | Built from keyword tools instead of customer language |
| Usability | Ecommerce and product | Unowned; the feed belongs to whoever built it in 2019 |
What works in most organizations isn’t a new team. It’s a named owner and a quarterly forum: one accountable person, a standing session with engineering, PR, content, ecommerce and legal, and a single dashboard all of them see. The forum exists because the failure mode isn’t incompetence — it’s that five levers sit in five reporting lines and nobody holds the whole picture.
Legal belongs in that room. The review rules, the liability developments, the transparency obligations — all live, and all cheaper to handle as constraints on the plan than as responses to an incident.
Seven questions for an agency
If you’re buying help, these separate the substantive from the performative.
What’s your evidence that this works? In writing.
How many runs per prompt per day does your measurement use, and do you publish confidence intervals?
Are you proposing anything that requires editing Wikipedia? If yes, end the conversation.
Are you proposing anything involving reviews, forum posts or testimonials we wouldn’t want printed in a trade publication? If yes, end the conversation.
What proportion of the work is on our own site? If it’s most of it, they’re selling you Chapter 6 and calling it Chapter 7.
What do you think llms.txt does? Cheap, reliable diagnostic.
What would make you tell us this isn’t working? Anyone who can’t answer has no falsifiable model.