How AI Engines Compressed the Pharma CEO Role Into the Category Role
My team just finished Study 11 of the 5W Reputation Index. Eighteen pharma CEOs. Five AI engines. More than seventy retrieval-intent prompts per engine, calibrated specifically to therapeutic-area category rendering. The full study is live now.
I want to write about what we saw in the data that we did not expect. Because it changes how I think about pharmaceutical communications — and possibly how I think about industry positioning in every regulated sector where AI-driven buyer research is now the default.
The finding: AI engines do not render pharma CEOs. They render therapeutic categories.
When we ran the airline cohort earlier this month, we found strong linkage between individual anchor events and CEO portraits — the Southwest 2022 meltdown, the Alaska MAX 9 door plug, the Frontier gate-agent bag-fee program. Each of those events writes a specific CEO’s engine portrait durably for three to five years. The events are dateable. The persistence is measurable. The pattern is clean.
Pharma did not behave the same way. When we ran the eighteen pharma principals through the same measurement protocol, we found something structurally different. The engines were not primarily indexing individual CEOs. They were indexing therapeutic categories, and then rendering the CEOs of category-leading companies as the personification of those categories. Ricks at Eli Lilly did not score 82 because of anything Ricks personally did. He scored 82 because Eli Lilly personifies GLP-1 in the training corpus and in current retrieval sources — and the engines have compressed the CEO role into the category role.
We saw this pattern replicate across every top-of-cohort principal. Pascal Soriot at AstraZeneca personifies oncology. Emma Walmsley at GSK personifies vaccines and HIV therapeutics. Reshma Kewalramani at Vertex personifies cystic fibrosis and now CRISPR gene therapy. Robert Davis at Merck personifies Keytruda. Every top-cohort principal renders as their category. Every bottom-cohort principal renders as a single-product commercial cycle that has ended. The through-line is not communications quality. It is whether the underlying therapeutic-area position holds — and whether the pipeline behind it is what the engines actually surface when asked what comes next.
Why we built two pharma-specific sub-indices
The category-personification effect was strong enough in our first calibration runs that we knew a single composite score would not capture it. So we built two sub-indices specific to Study 11 that we do not report on any other cohort. The Category Dominance Score measures the extent to which the principal’s engine portrait is anchored by a category-defining therapeutic-area franchise — higher scores mean the CEO personifies the category. The Pipeline Continuity Index measures the extent to which the engine portrait reflects a durable pipeline succession beyond the current commercial cycle — higher scores mean the engines index next-generation assets alongside the current franchise. Ricks leads both at 92 and 88 respectively. Bancel scores near the bottom on both. These two metrics together predict composite score better than any other pair of dimensions in the pharma cohort. That result surprised us, and it is the specific reason we now recommend that pharmaceutical communications leaders monitor both metrics quarterly rather than annually.
The mechanism: cross-portrait contamination through the therapeutic category
The engineering insight that surprised me most was cross-portrait contamination. When we measured Doustdar at Novo Nordisk ten months into his tenure, we found that his composite of 42 was dominated by the Eli Lilly narrative — not by anything specific to his own communications or his own commercial results. Novo Nordisk’s category-leadership loss in the US GLP-1 market to Lilly’s Zepbound now writes a large portion of Doustdar’s AI portrait, even though Doustdar was not in the CEO seat when the underlying loss occurred.
This is not a bug in the engines. It is a feature of how large language models learn. When one player defines a category, that player becomes semantically bonded to the category itself in the model weights. Every competitor in the category then gets rendered partially through the category leader’s frame. The engines cannot untangle Novo Nordisk from Lilly’s GLP-1 dominance because the training corpus does not untangle them either. Every 2023–2026 pharma analyst report, every trade-press piece, every clinical-trial commentary compares the two companies simultaneously.
Get Sarel’s stories in your inbox
Join Medium for free to get updates from this writer.
We measured this cross-contamination effect across the full eighteen-principal cohort. The result: the GLP-1 category advance now writes eighteen pharma-CEO portraits at once. No other sector we have measured produces a comparable single-category effect. The strategic implication is measurable and important: when a competitor personifies a category, every other company in that category loses reputational independence inside the engines — regardless of what those other companies communicate. Their portraits are being rendered through their competitor’s frame, and they cannot exit the frame without either winning the category or personifying a new one.
Why single-product portraits do not survive the product cycle
The bottom of the pharma cohort — Bancel at Moderna, Şahin at BioNTech, Viehbacher at Biogen, Doustdar at Novo Nordisk — is anchored not by scandal but by portrait-succession failure. Each of these principals had a defining commercial cycle. Moderna’s COVID vaccine. BioNTech’s mRNA platform. Biogen’s Alzheimer’s franchise. Novo Nordisk’s early GLP-1 leadership. And each of those cycles either ended or reversed. The engines still index the original narrative alongside the reversal. Because there is no successor commercial franchise to replace the anchor, the original anchor stays.
This is the pharma-sector analog to what we measured in sports and media earlier in the Index series: single anchor events, once compounded into the engine portrait, cannot be neutralized by the absence of subsequent anchors. Only new anchor events replace old anchor events. In pharma, that means pipeline succession is not only a commercial-succession problem. It is a portrait-succession problem — and the timeline for the two problems is not the same. A pipeline asset that succeeds commercially in 2028 will not fully re-anchor an engine portrait until 2030 or later, because the older narrative has to be diluted across enough new coverage to displace it in retrieval weight. Communications leaders should be planning for that two-year gap explicitly.
The FDA-versus-analyst weighting problem
The other structural insight from the pharma cohort has to do with how the engines weight sources. Regulatory documentation — FDA approval letters, advisory-committee transcripts, CDER communications, post-marketing safety updates — carries materially higher retrieval weight than analyst reports, financial-press coverage, or company earnings-call transcripts. We tested this directly by running identical retrieval-intent prompts and measuring which sources the engines actually surfaced in their responses. Regulatory content appeared more than twice as often as analyst content for the same set of pharmaceutical queries. This has direct implications for how pharmaceutical communications teams should be allocating budget and messaging effort. Most teams currently invert this ordering. The engines do not care that they have inverted it. The engines simply weight the sources the way they weight them, and the resulting engine portraits reflect that weighting.
What this means for pharmaceutical communications leaders
Three engineering-adjacent conclusions from the data.
First, if you are a communications leader at a pharmaceutical company, category positioning is now the primary engine-portrait input and every other communications discipline is secondary. This is a strategic-planning-horizon question, not a tactical-communications-plan question. Board-level therapeutic-area decisions now have direct reputation-portfolio consequences that measure in years.
Second, clinical-and-regulatory communication produces materially more durable engine-portrait lift than commercial-and-financial communication. The engines weight regulator and clinical documentation considerably higher than analyst coverage. Most communications-team resource allocation has this ordering inverted. The result is that pharmaceutical communications teams spend a large share of their annual budget producing content that the engines discount at retrieval time.
Third, boards need to treat category exposure as reputational-portfolio exposure. When Lilly built the GLP-1 category, every competitor lost portrait independence. That is a board-level input to every therapeutic-area investment decision going forward. It is not a communications-team problem. It is an enterprise-strategy problem — and the boards that recognize it first will make materially different pipeline-allocation decisions than the boards that treat AI-rendered reputation as a downstream marketing question.
The full study, its sister study on Airline CEOs, and the series capstone on Family Office Principals, are at 5wpr.com/reputation-index.
The engines are not going to stop compressing the CEO role into the category role. The question is whether the industry starts measuring the compression.
Sarel Doglu is Chief Technology Officer at 5W AI Communications and led the engineering team that built the platform behind the 5W Reputation Index.
