Efficiency has built holdco sameness. AI will make that an expensive choice.
The agencies that move first under LLM-mediated selection won’t be the ones with the cleanest schema markup. They’ll be the ones who have taken a clear, distinct position, writes Paul Evans.
Haymarket title Campaign recently ran an article with the headline “Clients are using LLMs for agency selection: Agencies urged to improve their own AI discovery, as clients increasingly use LLMs to write pitch lists.” The online reaction and discussion went straight to treating this as a technical discoverability problem – schema markup. citation placement. Generative engine optimisation.
That direction is sound and logical. But none of it answers the actual question the model is being asked or addresses the fundamental driver of discoverability. That question is the one an advertiser has always had. Why should I choose you? That driver is positioning.
Nobody set out to look identical
The holdco “sameness” I refer to in the title is in no way deliberate. But over time, there has been a race towards homogeneity through simplification. Collapse different operating brands into one. Share the trading desk. Standardise the client experience. Clean the P&L. Each move made sense on its own terms. Stacked together, they’ve been eliminating much of the difference and distinctiveness that customer-orientated positioning depends on.
Let’s look at some recent examples. WPP folded GroupM into WPP Media in 2025, with EssenceMediacom, Mindshare and Wavemaker reduced to client-service brands under a single P&L. In February this year, Cindy Rose’s Elevate28 plan went further, restructuring the entire group into four operating units and folding the major creative networks under WPP Creative. Rose told journalists WPP was “no longer a holding company.”
Omnicom closed its acquisition of IPG in November 2025 under the official line “Built for Intelligent Growth in the Next Era,” then announced it would retire DDB, FCB and MullenLowe by mid-2026 – three storied agency brands extinguished in a single line of an integration plan.
Publicis continues to run on the Power of One model as a self-described “Connecting Company for the Connected Age.” Dentsu is consolidating around One dentsu and Integrated Growth Solutions.
Read those four corporate descriptions side by side. Now imagine an LLM doing it, attempting to decipher any discernible difference between them.
Sameness has worked. For a time.
The status quo of sameness is seemingly palatable and profitable for every (human) stakeholder. Holdcos delivered the efficiency and simplicity that shareholders crave. Procurement and pitch consultancies got highly comparable proposals it could easily benchmark, leaving media trading and cost of service as the only line items that count. Sameness was a two-sided equilibrium. Agencies optimised for efficiency. Advertisers embraced the simplicity.
This is why “media agencies all sound the same” has been the most repeated and least acted on observation in our industry for 20 years. Human relationships have therefore largely determined agency pitch shortlists. It’s been the status quo that has worked. Until it hasn’t.
What changes when the buyer is a machine
An LLM is deliberate and logical – a pattern recognition engine – and will surface only what it is able to find and interpret. It will recommend where it sees difference in the context your business is discovered – and it’s already deciding whether to shortlist you.
Forrester’s 2025 Buyers’ Journey survey, published in January, found that 94% of B2B buyers use generative AI in their purchasing process, with AI search outranking vendor websites and sales reps as the most meaningful source of information.
When that AI-generated synthesis is applied to the four holdco statements above, it returns something close to a draw. The models then recommend from whatever signal it can find: citation volume, recency, third-party coverage – the discriminable substance that has been said about each agency beyond its own digital assets.
McKinsey research published last October found that a brand’s own site typically accounts for only five to 10% of the sources AI-powered search references; the rest come from affiliates, publishers, and user-generated content. Selection becomes an audit of who has been positioned in the world outside of your control.
Where AI learns about you
Muck Rack’s What Is AI Reading? study, now in its third edition, has tracked AI citation patterns across more than 25m links from ChatGPT, Claude and Gemini. The May 2026 edition reports that 84% of AI citations come from earned media. The figure has held between 82% and 89% across all three editions since July 2025. Owned channels barely register in the citation pool. This is what the execution of positioning means in the AI era.
Positioning – as I define it at V2RSION – is the context with which you are chosen. That “context” represents the components that make a business understood on the basis of a purchase decision to be made, grounded in a combination of present and future capability.
It has to hold six things at once: the customer problem being solved, the unique value created in solving it, the capability that makes that value real, the category that frames the choice, the alternatives the customer could choose, and the who that customer is.
It is this rich context, when read by LLMs in response to a specific question that carries intent, that will be returned as a specific recommendation.
The work no one wants to do
I argued in a Media Leader column earlier this year that the holdcos are “running from the word ‘agency’ at the precise moment AI is redefining it as the most valuable word the industry has”.
I still stand behind that argument as a source of differentiated context for any agency that wanted to get behind it fully and without compromise.
‘Agency’ – the word the holding groups are running from is the one they need most
What I’d add now is the cost of ignoring that opportunity – and the cost of having weak or “sameness” positioning generally.
The agencies that move first under LLM-mediated selection won’t be the ones with the cleanest schema markup. They’ll be the ones who have taken a position clear and distinct enough that a journalist, prospect client, pitch consultant – and now machine – can understand and prefer it.
But the market does not split into human and machine segments. It will split into agencies who have a context-rich position that clearly states why they should be chosen, and agencies that don’t. That’s the reality of positioning in the AI era.
Paul Evans is the founder and chief positioning engineer at V2RSION