Statistically I expect to lose most readers by the end of this paragraph. This is not a criticism of you the reader, nor of my writing talent or lack thereof, it's simply a function of the attenuated attention spans the present set of media interfaces encourage. What it does mean is I'd never recommend bottom of the page inventory, because I don't need a viewability metric to know most of the audience won't see that. That's a gut feeling though, and while I assume most of the readers, deep in the comforting folds of the pharma-advertising nexus who I work with, would agree with me.
No, gut is insufficient for full strategy, for full strategy we need data. So here's something interesting to chew on: how should Eli Lily's Ebglyss brand team change it's media mix as a result of the rapid growth of AI assistants and agents and that sort of thing? Been hearing a lot about it GEO's impact, so I thought I'd run some tests. My initial assumption was that GEO would cannibalize search, this does not appear to be the case.
Ebglyss is interesting- got the nod from the FDA for atopic derm in late 2024, taking a swing at everyone's favorite immuno-cure all, Dupixent. As with many drugs I've watched launched in the immunology world, it competes by being more targeted to the single indication and provider longer lasting efficacy. This is compelling compared to Dupixent because while effective, you have to take it twice as often, raising the odds of patient adherence problems. Since Dupixent does this effectively for a million conditions, it's often what docs reach for first when treating Atopic Dermatitis- if Ebglyss is good for that, Dupixent can work for the whole family, so why not try the buckshot solution. Indeed, market reports say maybe 30-40% of Ebglyss starts come after Dupixent, while I find the number of switchers from any relevant drug closer to 70% in our data.
Eczema, the umbrella AD falls under, is particularly especially germane for consumer marketing promotion because treatment for it is more elective, that is patient driven, than other conditions- a person can choose to suffer, as many do, rather than take on the manifold hurdles of the healthcare system. Hence the importance of consumer engagement- the patient is the chief advocate in looking for a cure, a doctor can at the end of the day say we're treating this sufficiently, it's not life threatening. We see this across conditions, I should note- while you can look at disease prevalence statistics from the CDC, the number of patients actually engaging with the system for care often falls far lower than that, perhaps at the rate of 50-60%, in any given year. We're all tired, those with higher disease burdens more tired than most, and believe me that shows up in utilization of elective and preventative care medicine.
This fact is incredibly important for pharma marketers constructing a strategically addressable market and a media plan, as you don't want your boss throwing out a number of potential patients twice the size of the real pool you can convert- that's a recipe for a short tenure. So what you want to do when peering into the crystal ball of population health statistics is understand a few truths and how those should relate to a plan for reaching the population:
1) how many people have this 2) how many are going to seek care 3) how can I nudge that number up with awareness through unbranded campaigns, testing, etc 4) how do I reach that population?
Enter, media mix – an analysis of how to reach those people, plus or minus the impact of broader awareness play that modulate the size of the total addressable pool. This is why I used Ebglyss- it's indication, atopic derm, is not a rare one, so it requires the tactics of endemic promotion which borrow heavily from CPG and other mass-media apparatus's.
One of the things that's intriguing when you look at media mix decisions is how close the winners and losers can be- just a few points one way or another can shift major marketing decisions, because plowing media into a truly omnichannel play without a good feedback loop is a sure recipe for failure. Without adequate data capture pipelines you'll be stuck with a static projection, of what looked like a good idea at one point in time, unable to respond to changes in audience composition which mean a given channel totally misses the mark or only gets a fraction of the population it should engage. One solution the media buying mechanisms, platforms, have tried to offer, is to enable buying everything under one roof. That's been partly successful you can buy display, mobile, ctv, and even DOOH and EHR under one roof in a few places now, but any credible strategy still entails the walled gardens of Meta, Google, and linear TV. A media mix plan that excludes these (search? social? tv?) is suicide, as you're guaranteed to only reach a fraction of who you need to. Let's say 60% of people with the condition are thinking about getting treated, if you shrink your pool again by just those reachable without those channels, you're reaching a fraction of a fraction of the relevant population. Don't blame me, blame media fragmentation. In summary, while many of the methods of buying ads have become increasingly frictionless, and measuring their performance within-channel has become frictionless, deciding between channels remains a labor-intensive effort of data wrangling and normalization across media ecosystems and structures. Said differently, we can now optimize within tactics extremely efficiently for near-term objectives, but strategic optimization at a brand strategy level remains a massive undertaking, particularly for endemic conditions obligated to pursue genuinely omnichannel initiatives by virtue of the scope of their markets.
Into this unstable concoction enters GEO (Generative Engine Optimization), an exceedingly broad bucket of tactics. I don't think there's a good definition for it yet, maybe a basic one would be "ways to make the category of AI tools people use mention your product". Does that include OpenAI's display advertising? I'm not even sure. What I am sure of is it's now competing for dollars alongside every other medium.
This is where the closeness of media mix decisions can trip you up, because the share GEO takes, while not dominant, is far from evenly distributed. My naive assumption was it'd all come from search – Google is ramming their LLM, Gemini, down everyone's throats in their search results, the "Google zero" phenomenon that appears to be absolutely brutal to publishers site traffic (credit to Nilay Patel at the verge for the term). In fact though, AI assistants don't erode competing channels gradually. They sit just below a handful of incumbents, and a small affinity increase is enough to flip a ranking and eject whoever was in third place.
So anyways I go to build my media mix model – I pull in a bunch of market research on media consumption trends split by demography. In this case we're talking a light complexity mix- nine channels, which we'll then analyze across a multilayer grid of four different demo attributes (want more detail? pay me). We then pull the patient cohort of atopic derm patients and feed it into this engine. What comes out is a per-demo recommendation of how to allocate budget across these. More sophisticated versions would bake in relative channel cost for efficiency but I didn't need that for this exercise.
The result surprised me: my conviction of GEO eating search didn't pan out. Let's think of LLM use and search as two categories of a bucket, query based channels. Search remains double the GEO allocation, 15.5% to 7.2%. Search hardly loses out, in fact when a query-based channel gains strength in this model, the channel that loses ground is consistently CTV, not Facebook, not Point-of-Care or Linear TV which reach audiences that aren't adopting it as rapidly. GEO and search are pulling from the same well when they compete for anything; that well is video-adjacent passive attention, not each other. If AI assistants aren't displacing search, where is their actual growth showing up? The clearest answer sits in the 30-44 age band, where AI Assistants and CTV/Streaming land neck and neck. Those are the channels which seem to be competing in growth in the same populations, are more substitutable for one another than search. In the longer term should the trend persist it seems marketers will need to decide between an ad on peacock vs openai than openai and google. Food for thought for the weekend.
Written purely by hand, as I've grown to resent internal accusations of AI overuse,
Ted
Originally published on LinkedIn.