Advanced melanoma in AI · Germany

Same Treatments. A Different Order.

We asked ChatGPT, Claude, Gemini and Perplexity the same 243 questions about the systemic treatment of advanced melanoma — in German, from the perspective of an oncologist in Germany — and measured which treatments each system named, and which it named first, across 972 answers.

Nivolumab-based treatment is named in 94% of answers and named first in 62%.

Asked about speed of response, AI names BRAF/MEK targeted therapy first in 74% of answers.

perspective: Oncologists
language: German
questions: 243
answers: 972
AI systems: 4
measured: 6 October 2026

01 · The treatment landscape

Nivolumab is named in 94% of answers, ipilimumab in 92%.

AI Visibility of the 10 tracked molecules, across all 972 answers or one system's 243.

How many answers name the molecule at least once?The active substance

The 10 tracked molecules, with the answers naming each at least once, overall and within each AI system. 972 responses scored, 243 per AI system.
Rank	Treatment	Tracked as	Linked	AI Visibility	Claude	ChatGPT	Perplexity	Gemini
1	Nivolumab	Molecule	—	94%913 / 972	82%199	98%238	100%242	96%234
2	Ipilimumab	Molecule	—	92%896 / 972	79%191	96%234	99%241	95%230
3	Pembrolizumab	Molecule	—	65%628 / 972	51%125	72%175	78%190	57%138
4	Dabrafenib	Molecule	—	61%596 / 972	53%129	53%130	70%170	69%167
5	Trametinib	Molecule	—	61%589 / 972	50%122	53%130	70%170	69%167
6	Encorafenib	Molecule	—	48%469 / 972	31%75	40%96	63%154	59%144
7	Binimetinib	Molecule	—	47%459 / 972	31%75	39%94	63%152	57%138
8	Vemurafenib	Molecule	—	41%396 / 972	41%99	34%83	58%142	30%72
9	Cobimetinib	Molecule	—	38%367 / 972	29%70	34%83	58%142	30%72
10	Relatlimab	Molecule	—	36%353 / 972	20%48	51%123	51%123	24%59

Bars are scaled to the highest value in the current view. Every figure is also written out.

02 · Prominence

Named is not the same as named first.

AI Visibility shows whether a treatment is named. Prominence shows where it appears in the order of tracked treatments named.

Each bar is all 972 answers: the filled length is AI Visibility; the parts show where the option first appears.

Named 1st
2nd
3rd
4th or later
Not named

Nivolumab-basednivolumab ± ipilimumab or relatlimab
62% named first94% named
Dabrafenib + trametinib
16% named first62% named
Pembrolizumab
16% named first65% named
Encorafenib + binimetinib
2.7% named first48% named
Vemurafenib + cobimetinib
1.1% named first41% named

Order of mention is not a preference, a recommendation or a clinical priority.

Encorafenib + binimetinib is named in 48% of answers, but named first in only 2.7%.

03 · The question changes the order

Asked about speed of response, AI mostly names BRAF/MEK therapy first. Asked about durability, it almost never does.

Which class of treatment is named first in answers to each kind of question, across the four AI systems combined. Each row states its own base.

BRAF/MEK targeted therapy named first
Immunotherapy named first
No tracked treatment named

Speed of response21 questions · 84 answers
Tumour response21 questions · 84 answers
Brain metastases24 questions · 96 answers
Safety & tolerability87 questions · 348 answers
Efficacy & survival69 questions · 276 answers
Durability of benefit21 questions · 84 answers

The contrast between speed of response and durability holds in each of the four AI systems.

From efficacy questions to safety questions

Pembrolizumab is named first in 3% of answers to efficacy questions and in 39% of answers to safety & tolerability questions.

Efficacy questions · 624 answers
Safety & tolerability questions · 348 answers

Nivolumab-based71% → 47%
Dabrafenib + trametinib21% → 8%
Pembrolizumab3% → 39%
Encorafenib + binimetinib2% → 4%
Vemurafenib + cobimetinib2% → 0%

When the question is about brain metastases

Dabrafenib + trametinib is named in 88% of answers about brain metastases; encorafenib + binimetinib in 35%.

All questions · 972 answers
Brain metastasis questions · 96 answers

Nivolumab-based94% → 97%
Dabrafenib + trametinib62% → 88%
Pembrolizumab65% → 16%
Encorafenib + binimetinib48% → 35%
Vemurafenib + cobimetinib41% → 19%

04 · BRAF status in the question

When BRAF V600 is in the question, the treatment order changes differently across AI systems.

Share of each system's answers that name BRAF/MEK targeted therapy first, by what the question says about BRAF status.

BRAF status not stated · 176 answers per system
BRAF V600 in the question · 55 answers per system

Claude14% → 75%24 → 41
ChatGPT9% → 16%15 → 9
Perplexity20% → 33%35 → 18
Gemini15% → 47%26 → 26

Which treatments are named

BRAF status not stated · 704 answers
BRAF V600 in the question · 220 answers

Nivolumab-based97% → 84%
Dabrafenib + trametinib56% → 92%
Pembrolizumab72% → 35%
Encorafenib + binimetinib42% → 80%
Vemurafenib + cobimetinib34% → 70%

In the 48 answers to the 12 questions that state BRAF wild-type — all of them about efficacy — a BRAF/MEK combination is named in 4 and is never named first.

05 · Across AI systems

Every AI system names nivolumab-based treatment first most often. Below it, they differ.

Named first: the share of each system's 243 answers in which the option is the first tracked treatment named.

Claude
ChatGPT
Perplexity
Gemini

Nivolumab-based56–67%
Dabrafenib + trametinib9–20%
Pembrolizumab6–33%33% in ChatGPT · 6% in Claude
Encorafenib + binimetinib0–5%
Vemurafenib + cobimetinib0–5%Named in 59% of Perplexity's answers, and never first there

Systems differ in how many treatment options they name per answer — on average 2.6 in Claude and 3.7 in Perplexity — so compare each system's pattern, not its level.

Measuring the order in one AI system describes one AI system.

06 · Molecule and brand

AI speaks in molecules.

of answers name at least one of the 10 tracked molecules: 98.5%
name at least one tracked brand: 0.4%

Nivolumab is named in 913 answers, Opdivo in 1. Of all 972 answers, 4 name any of the 11 tracked brands.

Beyond Visibility

Visibility and Prominence show where a treatment appears. Not what AI says about it.

The next layer is how AI represents the therapy: how efficacy and safety are characterised, whether important qualifications survive, how it is compared with competitors, and which sources underpin the answer.

Full assessment

What a full assessment reveals

Example: Your brandGermany · Oncologists

Competitive positioning
Product attributes
Strategic alignment
Source intelligence
Content recommendations

The free analysis shows which names appear, and in what order. Theralexis shows how AI models describe your brand, what shapes that description, and what to do about it through source and content recommendations.

About this benchmark

What was measured

One benchmark run: ChatGPT, Claude, Gemini and Perplexity, each with web access, answered the same questions about the systemic treatment of advanced melanoma — in German, from the perspective of an oncologist in Germany — once each, on 6 October 2026.

AI Visibility — the share of answers in scope that name a treatment at least once.

Prominence — where a treatment first appears in the order of the tracked treatments an answer names. Named first is the share of all answers in scope, including those naming no tracked treatment, in which a treatment option comes first.

Treatment options — combination partners count as one option, placed where either partner first appears. Nivolumab-based covers nivolumab alone or with ipilimumab or relatlimab. Other treatments are not part of the order.

Question contexts — a Theralexis grouping, every question in exactly one. They differ in size, so each figure states its base. Efficacy questions are every context except safety & tolerability.

BRAF status — taken from the question's wording. Biomarkers named in the answers are not reported, nor are further tracked molecules, including chemotherapies.

What this does not show

Order of mention is not a preference, a recommendation or a clinical priority, and being named says nothing about clinical correctness. The benchmark does not show how a treatment is described, which sources shaped the answer, or why one treatment comes before another — the order can reflect the question, how a system structures a list, or the evidence it draws on. Systems differ in how many treatments they name per answer. One run in one market, language and audience; results may differ elsewhere or over time.

Full benchmark

Explore the full benchmark.

The full benchmark goes deeper — every treatment option by question and AI system, and where the order changes.

Schedule a meeting