In AI search, a 30-year-old drug can outrank this year's blockbuster. Here is why — and what the data says about how AI models distribute brand mentions across therapeutic categories.

Share of voice has always been the currency of competitive intelligence in pharma. In traditional media and search, it was largely a function of spend: bigger budgets bought more impressions, higher rankings, and louder presence. But AI search operates on entirely different rules. When a patient asks ChatGPT about Crohn's disease treatments, or a physician queries Perplexity for psoriasis drug comparisons, the AI model does not consult an ad server. It consults its training data, its source ecosystem, and its understanding of clinical evidence.

The result is a share of voice distribution that no marketing budget can buy — and one that systematically favors drugs with the deepest evidence base, not the newest approval date.

We used the [PharmaGEO platform's agnostic mode](#) to map share of voice across three major therapeutic areas: Crohn's disease, psoriasis, and atopic dermatitis. The findings reveal a competitive landscape that should change how both legacy brands and new launches think about AI visibility.

Data source: PharmaGEO platform agnostic mode analysis across OpenAI, Gemini, and Perplexity (2025).

How AI Distributes Share of Voice — and Why It Differs from Traditional Search

In traditional search, share of voice is a paid and earned metric. Brands that invest in SEO, SEM, and content marketing earn more visibility. Position one on Google captures roughly 30% of clicks, and the top three results take more than 50%. It is a winner-takes-most system.

AI search inverts this dynamic.

When a generative AI model responds to a therapeutic query, it synthesizes information across its entire source ecosystem. It does not rank brands sequentially. Instead, it mentions brands within a narrative answer, weaving multiple treatments into a single response. This fundamentally changes the distribution pattern.

Instead of one brand dominating position one, AI spreads mentions across many brands within a narrow range. Our data shows that in every therapeutic area we analyzed, the top five to six brands were separated by no more than 1 to 2 percentage points of share of voice.

This is not how traditional search works. This is something new.

Key Takeaway: AI search does not produce a single winner per query. It produces a distributed narrative where multiple brands share near-equal mention rates — and the brands included in that narrative are determined by evidence depth, not marketing spend.

The Data: Three Therapeutic Areas Mapped

Crohn's Disease (French-Language Queries)

Brand Share of Voice Drug Class
Imurel (azathioprine) 10% Immunosuppressant (legacy)
Remicade (infliximab) 9% Anti-TNF biologic
Humira (adalimumab) 9% Anti-TNF biologic
Stelara (ustekinumab) 9% Anti-IL-12/23 biologic
Entyvio (vedolizumab) 9% Anti-integrin biologic

Five brands within a single percentage point. The most striking finding: Imurel — a conventional immunosuppressant first used decades ago — holds the highest share of voice in AI-generated responses about Crohn's disease. Not Humira, the best-selling drug in pharmaceutical history. Not Stelara or Entyvio, which represent the current standard of care for many patients. Imurel.

This is the legacy advantage in action. Azathioprine has been studied in Crohn's disease for over 40 years. It appears in every major treatment guideline, thousands of published studies, and decades of real-world evidence reports. AI models trained on this literature cannot avoid it — and they do not.

Grounding rate: 94% (135 sources cited). Nearly every AI-generated claim about Crohn's treatments links to a verifiable source.

Psoriasis (English-Language Queries)

Brand Share of Voice Drug Class
Remicade (infliximab) 8% Anti-TNF biologic
Taltz (ixekizumab) 8% Anti-IL-17 biologic
Cosentyx (secukinumab) 8% Anti-IL-17 biologic
Otezla (apremilast) 7% PDE4 inhibitor (oral)
Humira (adalimumab) 7% Anti-TNF biologic
Stelara (ustekinumab) 7% Anti-IL-12/23 biologic

Six brands within a single percentage point. Psoriasis shows the most fragmented share of voice of any therapeutic area in our analysis. No single brand reaches even 10%. The leading position is shared between Remicade (a legacy biologic first approved in the late 1990s), and two newer IL-17 inhibitors, Taltz and Cosentyx.

The presence of Remicade at the top is another demonstration of the legacy effect. In clinical practice, Remicade is no longer the first-line biologic for most psoriasis patients. But in AI-generated responses, its decades of published evidence give it a weight that current market share alone would not predict.

Grounding rate: 96% (224 sources cited). Psoriasis has the deepest source pool in our dataset, which likely contributes to its highly fragmented distribution — more sources mean more brands represented.

Atopic Dermatitis (English-Language Queries)

Brand Share of Voice Drug Class
Cibinqo (abrocitinib) 11% JAK inhibitor (oral)
Dupixent (dupilumab) 10% Anti-IL-4/13 biologic
Rinvoq (upadacitinib) 10% JAK inhibitor (oral)
Adbry (tralokinumab) 10% Anti-IL-13 biologic

Four brands within 1 percentage point. Atopic dermatitis represents the newest drug class battlefield, with JAK inhibitors and targeted biologics competing head-to-head. Here, Cibinqo edges out Dupixent — a result that would surprise most brand teams given Dupixent's dominant market position.

Why? Cibinqo, as a newer JAK inhibitor, generated intense publication activity around its approval. Comparative studies, safety analyses, and positioning papers flooded the literature. AI models, trained on this concentrated burst of evidence, gave it proportional weight.

Grounding rate: 100% (97 sources cited). Every single AI-generated claim in the atopic dermatitis dataset links to a verifiable source. This is the highest grounding rate in our entire analysis.

Key Takeaway: Across all three therapeutic areas, no single brand dominates AI share of voice. The distribution is remarkably flat, with top brands separated by 1 to 2 percentage points. Legacy drugs with deep evidence bases consistently match or exceed newer, higher-revenue competitors.

The Legacy Advantage: Why Older Drugs Win in AI

The pattern is consistent across every pathology we analyzed. Legacy drugs — those with 10, 20, or 30+ years of published evidence — maintain a share of voice in AI search that exceeds what their current market position would predict.

Three mechanisms drive this:

1. Publication volume compounds over time. A drug approved in 2000 has had 25 years to accumulate clinical trials, meta-analyses, guidelines, case reports, and real-world evidence studies. Each publication adds another node to the source ecosystem that AI models draw from. A drug approved in 2023 may have stronger clinical data, but it has far fewer total publications.

2. AI models weight breadth of evidence, not recency. Unlike a Google algorithm that can be tuned to prioritize recent results, large language models synthesize across their entire training corpus. A drug mentioned in 500 publications across 20 years carries aggregate weight that a drug mentioned in 50 publications across 2 years cannot match — regardless of how strong those 50 publications are.

3. Guideline inclusion is a force multiplier. Legacy drugs appear in multiple generations of treatment guidelines. Every guideline that mentions a drug becomes a high-authority source in the AI's evidence hierarchy. Newer drugs may appear in the most current guidelines, but legacy drugs appear in current and historical guidelines, multiplying their source presence.

Key Takeaway: AI models favor depth of historical evidence over recency. Legacy brands with decades of publications, guidelines, and real-world data maintain higher share of voice than their current market dominance might suggest. This is not a bug — it is how evidence synthesis works at scale.

Co-Mention Networks: How AI Groups Your Competitors

Share of voice tells you how often a brand is mentioned. Co-mention networks tell you which brands are mentioned together — revealing how AI models naturally frame competitive comparisons.

This is competitive intelligence that traditional media monitoring cannot provide.

Crohn's Disease: The Biologic Triad

The strongest co-mention pairs in Crohn's disease form a clear pattern:

- Remicade <> Humira — the original anti-TNF comparison

- Remicade <> Stelara — TNF vs. IL-12/23 mechanism comparison

- Humira <> Stelara — the two most-prescribed biologics linked together

This triad reflects how clinical literature has historically compared treatments in Crohn's disease. AI models reproduce this framing because the source ecosystem is built around these comparisons. For a newer drug trying to break into this competitive frame, the barrier is not clinical superiority. The barrier is co-mention inertia — the entrenched pattern of which drugs AI models associate with each other.

Psoriasis: The Oral vs. Biologic Frame

In psoriasis, the dominant co-mention pattern shifts:

- Otezla <> Humira — oral PDE4 inhibitor vs. injectable biologic

This pairing is the primary competitive frame AI models use when discussing psoriasis treatment. It reflects a clinical reality — patients and physicians frequently weigh the convenience of oral therapy against the efficacy of biologics — but it also reveals that AI models organize competitive comparisons around mechanism-of-action contrasts, not brand-level rivalry.

For brands like Taltz and Cosentyx that compete primarily with each other (both IL-17 inhibitors), this framing is a challenge. AI models are more likely to compare across classes than within them.

Atopic Dermatitis: Two Distinct Clusters

Atopic dermatitis reveals the most complex co-mention network:

Cluster 1 — JAK inhibitors and biologics:

- Rinvoq <> Adbry — next-generation competition

Cluster 2 — Topical therapies:

- Protopic <> Elidel <> Eucrisa <> Opzelura — calcineurin inhibitors and PDE4/JAK topical agents grouped together

AI models in atopic dermatitis create two separate competitive arenas: a systemic therapy conversation and a topical therapy conversation. Brands that exist in one cluster are rarely co-mentioned with brands in the other. This has significant implications for positioning strategy — a systemic therapy competing against topicals needs a different AI content strategy than one competing against other systemics.

Key Takeaway: Co-mention patterns reveal how AI naturally frames competitive comparisons. These frames are driven by historical publication patterns and mechanism-of-action groupings, not by current market dynamics. Brands that do not appear in the relevant co-mention cluster are effectively excluded from the AI-generated competitive conversation.

The Grounding Floor: Why Sources Are Everything

One of the most significant findings in this analysis is the grounding rate — the percentage of AI-generated claims that link to a verifiable source.

Therapeutic Area Grounding Rate Sources Cited
Crohn's Disease 94% 135
Psoriasis 96% 224
Atopic Dermatitis 100% 97

These numbers are extraordinary. Across all three therapeutic areas, AI models ground nearly every factual claim in a cited source. This means that AI-generated drug information is not hallucinated — it is synthesized from a defined source ecosystem.

The implication is both reassuring and alarming.

Reassuring: AI models are not fabricating drug information. They are drawing from real clinical literature, guidelines, and institutional sources. The reliability concerns documented in our broader analysis are real, but they stem from synthesis errors and omissions, not wholesale invention.

Alarming: If grounding rates are 94-100%, then the source ecosystem entirely determines what AI says about your drug. Brands that are not well-represented in the indexed source ecosystem — clinical literature, treatment guidelines, regulatory documents, institutional content — are essentially invisible. You cannot be mentioned if you are not in the sources that AI models rely on.

This creates a structural barrier for new launches. A drug approved six months ago has a fraction of the source footprint of a drug with 20 years of publications. And with grounding rates this high, AI models will not speculate about drugs they lack source evidence for. They will simply omit them.

Key Takeaway: Grounding rates of 94-100% mean AI models are almost entirely source-dependent when discussing drug therapies. If your brand is not in the source ecosystem — the clinical literature, guidelines, and institutional content that AI models index — you are invisible. There is no paid shortcut around this.

Implications for New Launches

The data presents a clear challenge for drugs entering the market. If AI models favor evidence depth and source volume, how does a new brand compete?

1. Front-load the evidence ecosystem. The publication strategy for a new drug must account for AI visibility as a distribution channel. Peer-reviewed publications, real-world evidence studies, and guideline inclusion are not just academic exercises. They are the raw material AI models use to determine whether your drug exists in a therapeutic conversation.

2. Target co-mention networks deliberately. New drugs need to engineer their way into existing co-mention patterns. This means publishing head-to-head comparisons, positioning papers, and mechanism-of-action reviews that place the new brand alongside established competitors. If the literature does not compare your drug to the incumbents, AI models will not make that comparison either.

3. Prioritize structured, authoritative content. AI models weight institutional and regulatory sources heavily. Ensuring your drug has comprehensive representation on regulatory authority websites, major medical institution pages, and clinical guideline databases creates high-authority source nodes that can punch above their weight relative to their recency.

4. Monitor AI share of voice from day one. Most new drug launches track traditional media share of voice from launch. AI share of voice should receive equal attention. Use [agnostic mode analysis](#) to track where your brand sits relative to incumbents and adjust your content strategy accordingly.

Implications for Established Brands

For legacy drugs, the data is not simply good news. It is a strategic asset that requires active management.

1. Your evidence depth is a moat — maintain it. The share of voice advantage that legacy drugs hold in AI is directly proportional to their ongoing source presence. Brands that stop publishing, stop pursuing guideline inclusion, or reduce their medical affairs footprint will see their AI share of voice erode as newer competitors accumulate evidence.

2. Monitor co-mention positioning. Being co-mentioned with the right competitors reinforces your position. Being co-mentioned with the wrong competitors can reframe your brand unfavorably. Track your co-mention network to ensure AI models position you within the competitive frame that best serves your strategy.

3. Address the accuracy gap. Legacy drugs with deep evidence bases sometimes suffer from AI models citing outdated information. A drug with 30 years of literature will have some publications that reflect superseded data — old dosing recommendations, discontinued indications, or adverse event profiles that have been refined by subsequent research. Actively ensure that the most current evidence is prominent and well-structured in the source ecosystem to prevent AI from surfacing outdated claims. Leverage pharmacovigilance monitoring strategies to catch and correct these inaccuracies.

4. Leverage the democratization effect. In traditional search, smaller brands struggle to compete with blockbuster budgets. In AI search, the playing field is more level. A legacy drug with a strong evidence base but a modest marketing budget can maintain share of voice parity with competitors that outspend it tenfold in traditional channels. AI visibility is one area where evidence quality matters more than spend.

Key Takeaway: For legacy brands, AI share of voice is a defensive asset. The evidence depth that took decades to build is now a competitive moat — but only if it is actively maintained and updated. For new launches, the path to AI visibility runs through the source ecosystem, not the ad budget.

Frequently Asked Questions

What is share of voice in AI search for pharma brands?

Share of voice in AI search measures how frequently a brand is mentioned when an AI model responds to therapeutic queries. Unlike traditional search where SoV tracks rankings and impressions, AI share of voice tracks brand mention frequency within narrative answers. In our analysis, we found that AI distributes mentions much more evenly than traditional search — top brands in any therapeutic area are separated by just 1 to 2 percentage points.

Why do legacy drugs have higher share of voice than newer drugs in AI?

AI models synthesize information across their entire training data, which includes decades of clinical literature. Legacy drugs benefit from publication volume, guideline inclusion across multiple generations, and long-term real-world evidence — all of which create a larger source footprint. A drug with 30 years of publications has more source nodes than a drug with 3 years, and AI models weight this cumulative evidence proportionally.

How does AI share of voice differ from traditional media share of voice?

Traditional share of voice is largely driven by paid and earned media: advertising spend, PR coverage, SEO investment, and content marketing. AI share of voice is driven by the source ecosystem — clinical publications, treatment guidelines, regulatory documents, and institutional content. You cannot buy a higher mention rate in AI responses. It must be earned through evidence and source presence.

Can pharmaceutical brands influence their AI share of voice?

Yes, but not through traditional marketing channels. Brands can influence their AI share of voice by building a robust source ecosystem: publishing peer-reviewed research, ensuring guideline inclusion, maintaining comprehensive regulatory documentation, and creating structured institutional content. Co-mention positioning can also be influenced by publishing comparative studies that place your brand alongside key competitors.

What are co-mention networks and why do they matter?

Co-mention networks map which brands AI models mention together in the same response. These patterns reveal how AI frames competitive comparisons. For example, in psoriasis, Otezla and Humira are frequently co-mentioned because AI frames the oral-vs-biologic comparison as the primary competitive question. Brands that do not appear in relevant co-mention clusters are excluded from the competitive conversation AI models present to patients and physicians.

How often should pharma teams monitor AI share of voice?

Given that AI models update their training data and retrieval sources regularly, monthly monitoring is the recommended minimum frequency. Major model updates or significant competitive events (new drug approvals, guideline revisions, major publication releases) should trigger immediate re-analysis. Use a standardized agnostic mode framework to ensure consistent measurement across time periods and AI platforms.

Conclusion: The Evidence Economy

The data is unambiguous. In AI-generated therapeutic conversations, share of voice is not a function of brand awareness, marketing spend, or even current clinical relevance. It is a function of evidence depth — the total volume and authority of the source ecosystem a brand has built over time.

This creates a landscape with three defining characteristics:

Distribution is remarkably flat. No single brand dominates any therapeutic area. Five to six brands share near-equal mention rates within 1 to 2 percentage points. AI search is not a winner-takes-most system. It is a coalition of evidence.

Legacy drugs punch above their market weight. Imurel leads share of voice in Crohn's disease. Remicade leads in psoriasis. These are not the highest-revenue brands in their respective categories, but they are the most deeply evidenced. AI models do not know which drug sells the most. They know which drug has the most published evidence.

Sources are the only currency that matters. With grounding rates between 94% and 100%, AI models are almost entirely dependent on their source ecosystem. Brands outside that ecosystem do not exist in AI-generated responses. There is no amount of promotional activity that can substitute for presence in the clinical literature, treatment guidelines, and regulatory databases that AI models index.

For pharmaceutical brand teams, the strategic implication is clear. The investment that drives AI visibility is not digital advertising. It is medical affairs. It is publication strategy. It is guideline engagement. It is the slow, deliberate work of building an evidence base that AI models cannot ignore.

The brands that understood this five years ago are already reaping the rewards. The brands that understand it today still have time. The brands that wait will find themselves asking why a 30-year-old immunosuppressant is outperforming their billion-dollar biologic in the one channel that patients and physicians are increasingly turning to first.

Data source: PharmaGEO platform agnostic mode analysis across OpenAI, Gemini, and Perplexity (2025).

This article is part of the PharmaGEO Insights Series on AI visibility for pharmaceutical brands. Related reading: The Brand Recognition Crisis in AI | The Reliability Tax | Benchmarking 23 Brands on GEO Pharma | The Pharmacovigilance Blind Spot | AI Model Comparison

See how your brand appears in AI answers.

Get a cross-LLM reputation report in minutes. No patient data. EU-based storage.

Book a demo Request a sample report

Data source: PharmaGEO platform analysis of 23 pharmaceutical brands across OpenAI, Gemini, and Perplexity (2025)