One brand, three languages, three different realities
Adtralza (tralokinumab) is an approved biologic for atopic dermatitis. In the May 2026 PharmaGEO public index, it earned a 13.8% Answer Rate (AR) in English-language queries on OpenAI — ranking ninth in a ten-brand field. Scroll to the French-language results for the same query set, same engine, same week: AR climbs to 48.8%, rank four. Switch to Spanish: 51.9% AR, rank four again.
That is a 38-percentage-point swing between English and Spanish on a single brand. Not a different product. Not a different indication. The same molecule, two brand namespaces, two completely different AI footprints.
This is the central finding of the language-geography analysis in the May 2026 PharmaGEO public index, and it has direct operational implications for every pharma team running a GEO program exclusively in English.
Why the gap exists
Tralokinumab was approved in the EU under the name Adtralza before it reached the US market, and its clinical development, EPAR filings, regulatory timelines, and disease-state coverage skewed heavily toward European sources in the years before a US launch. The non-English internet — specifically French- and Spanish-language medical society publications, European clinical registries, and EU regulatory documents — accumulated a richer, denser citation environment for Adtralza years before English-language content caught up.
LLMs are not translating content; they are retrieving from language-specific source pools. The French and Spanish pools happen to contain far more high-authority tralokinumab content than the English pool does. The AI answer reflects that asymmetry faithfully. A brand's AI visibility is regulated by the language of its approval geography, not by its global media spend or its English-language digital footprint.
The Adtralza numbers in full
The complete Answer Rate comparison across all three languages, drawn from the May 2026 PharmaGEO public index (OpenAI, Atopic Dermatitis query set):
| Brand | English AR | English Rank | French AR | French Rank | Spanish AR | Spanish Rank | Max Language Swing |
|---|---|---|---|---|---|---|---|
| Dupixent | 65.5% | #1 | 63.7% | #1 | 62.3% | #1 | 3.2 pp |
| Rinvoq | 59.8% | #2 | 57.5% | #2 | 59.7% | #2 | 2.3 pp |
| Cibinqo | 58.6% | #3 | 56.3% | #3 | 58.4% | #3 | 2.3 pp |
| Adtralza | 13.8% | #9 | 48.8% | #4 | 51.9% | #4 | +38.1 pp |
Source: May 2026 PharmaGEO public index, OpenAI engine, Atopic Dermatitis query set. AR = Answer Rate (share of prompts in which the brand is mentioned).
The stable tier and what it tells you
The Adtralza finding would be less significant if every brand moved unpredictably across languages. But the data shows a clear two-tier structure.
Tier one: globally launched brands barely move
Dupixent, Rinvoq, and Cibinqo — all with simultaneous or near-simultaneous FDA and EMA approvals, followed by years of global clinical publication and multilingual society coverage — show Answer Rate variances of less than 3.2 percentage points across all three languages. Their AI footprints are language-stable because their real-world clinical literature footprints are language-stable.
Tier two: regionally approved brands are wildly variable
Brands with asymmetric approval timelines show multi-decile Answer Rate swings when the query language shifts. Adtralza is the most extreme example in the May 2026 index, but the pattern applies to any brand whose approval geography did not match its primary commercial market.
If you have not benchmarked your brand's AI visibility in each target-market language separately, you have no idea what story AI is telling about you in those markets.
Brand-name geography: the Ebglyss anomaly
The language divergence is not only a question of Answer Rate magnitude. It is also a question of which brands appear on the list at all.
| Rank | English Top 10 | French Top 10 | Note |
|---|---|---|---|
| #1 | Dupixent | Dupixent | Stable leader |
| #2 | Rinvoq | Rinvoq | Stable |
| #3 | Cibinqo | Cibinqo | Stable |
| #4 | — | Adtralza | Not ranked in EN top 10 |
| #8 | — | Ebglyss (20.0%) | Absent from English top 10 |
| #10 | Olumiant (10.3%) | — | Drops out of FR top 10 |
Source: May 2026 PharmaGEO public index, OpenAI engine, Atopic Dermatitis query set. Rankings abbreviated for illustration; full 10-brand tables available in the index.
Ebglyss (lebrikizumab — commercialized by Almirall in the EU under the Ebglyss name; by Eli Lilly as Adbry in the US) ranks eighth in the French Top 10 with a 20.0% Answer Rate. It does not appear in the English Top 10 at all. The English top-10 closes at Olumiant with 10.3% AR; Ebglyss falls below that threshold in the English source pool.
This is brand-name geography in practice. LLMs do not recognize Adbry and Ebglyss as the same molecule when processing language-native content. French-language medical sources reference the EU brand name. English-language US sources reference the US brand name. The retrieval engine surfaces what its source language contains, and the two brand entities accumulate separate visibility scores in separate language pools.
The same molecule, two AI identities
For a brand team managing both Adbry and Ebglyss — or any product with distinct regional brand names — this creates a fragmented GEO challenge. Optimizing Adbry content in English does nothing to improve Ebglyss visibility in French. The EU-name brand and the US-name brand must be treated as independent entities in their respective language GEO programs. Most teams are not staffed or structured to do this, because they have been measuring AI visibility only in one language.
Localization is not translation
A translated Spanish-language brand page competes in the Spanish-language source pool — a pool that already weights certain brands, clinical narratives, and regulatory documents differently from the English pool. Inserting a translation does not change what that pool already contains. It adds one data point to a source environment shaped by years of Spanish-language clinical publication, EU regulatory documents, and regional medical society guidance.
Real non-English GEO requires original content anchored to what the language-native source pool already values: the society guidelines, regulatory bodies, and publication venues that carry authority in French or Spanish medical retrieval — not English-first content run through a translation workflow.
EMA EPARs: an underused citation lever
The May 2026 PharmaGEO public index reveals one structural advantage brands have in non-English markets: EMA European Public Assessment Reports (EPARs) are already functioning as citation vectors even in English-language answers. EMA EPAR documents appear five times in the Atopic Dermatitis English-language top-10 citation set and twice in the Psoriasis top-10 — a material citation frequency for a single regulatory domain at ema.europa.eu.
If EMA EPARs are already pulling weight in English-language retrieval, their influence in French- and Spanish-language retrieval — where EU regulatory documents carry even greater native authority — is plausibly larger still. Brands with EPARs in place have a regulatory citation asset that most English-only GEO programs are not actively leveraging. Making EPAR content accessible, structured, and linked is a low-effort, high-authority citation move that serves both EU-language and English GEO simultaneously.
The scale context: why this matters now
The reason language-level GEO divergences warrant urgent attention is the scale of the audience on the other side of the AI answer layer. Spectrum Science estimates 230 million weekly health questions are asked of ChatGPT alone. That volume is not US-centric or English-only — it spans every language in which ChatGPT is active, which is most of them.
Meanwhile, IQVIA data from March 2026 reports that 54% of HCPs are using generative AI in clinical contexts. European HCPs querying in French, German, Spanish, or Italian are drawing on language-native source pools. The answer they receive about a drug — its efficacy, its positioning relative to competitors, which brand name the AI uses — is shaped by the language-geography dynamics described in this article.
A brand team measuring only English-language AI visibility is measuring a fraction of the audience and potentially misreading its own competitive position in its most important commercial markets.
What to do about it
1. Run separate GEO scores per language
Answer Rate and Share of Voice must be benchmarked independently for each language in which your brand has commercial activity. An aggregate or English-proxy score will mask the Adtralza pattern — a brand that appears healthy in English may be underperforming badly in the languages where its clinical content base is thinner, or dramatically outperforming in languages where its EU approval history built a richer source pool. Neither situation is visible without language-level measurement.
The minimum viable language set for any EU-active brand is English, French, and Spanish. For broader EU portfolios, German and Italian are worth adding. Each language scores against its own source pool and produces its own competitive ranking.
2. Audit the non-English brand-name landscape
If your molecule has different brand names in different geographies — whether because of parallel licensing arrangements, regional commercialization partners, or distinct approval pathways — those brand names must be treated as separate GEO entities. Map every brand-name variant your molecule carries across markets. Run separate Answer Rate queries using each name in its native-language context.
Any product with regional naming conventions carries the same risk: the EU-name brand and the US-name brand accumulate separate AI identities, and optimizing one does nothing for the other.
3. Build language-native content, not translated content
Identify the high-authority source types in each target language pool. For French, this means European clinical guidelines, French medical society publications, and EMA regulatory documents. For Spanish, it includes AEMPS regulatory content and Ibero-American medical association resources. Native content integrates into these pools more effectively than translated English originals.
4. Activate EMA EPAR content as a cross-language citation asset
EMA EPARs are already generating citations in English-language retrieval. Ensure your EPAR documents are fully accessible, that brand-page content links to the relevant EPAR, and that the EPAR's clinical language is reflected in your own disease-state and prescribing information pages. The same EPAR that boosts English-language citation rate will carry even more weight in French and Spanish retrieval, where EU regulatory documents are the primary authoritative source type.
The strategic frame
GEO programs built on one language are making the same mistake early SEO programs made when they assumed one country's strategy would transfer globally. Language-geography divergence is a structural feature of how LLMs work: they retrieve from language-native source pools shaped by decades of region-specific clinical development, regulatory activity, and medical publication. The AI answer mirrors the source pool it draws from.
The brands that win AI visibility in European markets are not necessarily the ones with the largest English digital footprint. They are the ones that built high-authority source presence in the languages where their most important HCP audiences ask questions. For many EU brands, that work is partially done already — in EPARs, EU clinical publications, and regional society relationships. The GEO work is making that existing content structured, accessible, and correctly linked so retrieval engines can process it.
The language geography of AI visibility is not a future risk to model. It is a present-tense measurement gap. The HCPs asking questions in French and Spanish are receiving AI answers shaped by these dynamics right now, whether or not any brand team is watching.
Want a real audit on your brand across languages? Request a sample report or get the full PharmaGEO Playbook.