About this benchmarkRheumatoid arthritis · USRheumatologists · USAI answers generated with a US rheumatologist persona4 AI systemsPre-ACR baseline, 30 October 2026
AI Representation · Rheumatoid Arthritis · US
Pre-ACR baseline: 1,816 answers to 454 open questions and 960 answers to 240 comparative questions from four web-enabled AI systems, collected 30 Oct 2026. Not efficacy, safety or clinical preference.
01Representation
AI speaks molecule, not brand: 6.3% of answers name any in-scope brand.
43 in-scope brands. Brand naming is not preference.
6.3%115 of 1,816 answers name an in-scope brand
Market · US
AI Visibility & Representation across four AI models
All
Claude
ChatGPT
Perplexity
Gemini
02Competition
Upadacitinib is 7th for visibility, yet AI favours it over adalimumab in 23 of 32 head-to-head answers; adalimumab is favoured in none. 8 answers are judged equal, all on cardiovascular or infection questions, and 1 is conditional.
SELECT-COMPARE is the head-to-head trial the AI answers cite. This describes AI answers, not a clinical recommendation.
AI Visibility
Their 32 head-to-head answers
23Upadacitinib favoured
0Adalimumab favoured
03Context
No JAK inhibitor is clearly favoured in any of 64 cardiovascular or thromboembolic comparisons, in each of the four AI systems.
Cardiovascular and thromboembolic risk: no JAK inhibitor is favoured in any comparison; tocilizumab is favoured in 8 of 16, abatacept in 6 of 12 and sarilumab in 6 of 12.
JAK inhibitor vs non-JAK therapy, by question type
Efficacy & speed116 judgments
52 · 45%Equal 39 · Conditional 4 · Other 174 · 3%
Convenience & administration20 judgments
18 · 90%Equal 1 · Conditional 0 · Other 10 · 0%
Infection & zoster36 judgments
0 · 0%Equal 11 · Conditional 8 · Other 710 · 28%
CV / VTE risk64 judgments
0 · 0%Equal 5 · Conditional 22 · Other 532 · 50%
Therapy lens
Choose a therapy to see how AI answers represent it on every measure this benchmark publishes. Each measure has its own base, and they are never added together.
AI Visibility
Is it named at all?
38%
697 of 1,816 answers · rank 2 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
9%
130 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
37 · 22%Favoured
25 · 15%Comparator favoured
168 judgments against 12 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapyFavoured in 1 of 12
After TNF-inhibitor failureFavoured in 3 of 28
Pregnancy planning
Not shown. Compared fewer than 8 times in this profile, or not at all.
Cardiovascular & thromboembolic riskFavoured in 6 of 12
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Abatacept head to head →Follow Abatacept across the patient profiles →
AI Visibility
Is it named at all?
35%
644 of 1,816 answers · rank 3 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
8%
117 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
19 · 20%Favoured
8 · 8%Comparator favoured
96 judgments against 11 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapy
Not shown. Compared fewer than 8 times in this profile, or not at all.
After TNF-inhibitor failureFavoured in 14 of 48
Pregnancy planning
Not shown. Compared fewer than 8 times in this profile, or not at all.
Cardiovascular & thromboembolic risk
Not shown. Compared fewer than 8 times in this profile, or not at all.
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Rituximab head to head →Follow Rituximab across the patient profiles →
AI Visibility
Is it named at all?
28%
508 of 1,816 answers · rank 4 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
7%
99 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
59 · 37%Favoured
7 · 4%Comparator favoured
160 judgments against 13 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapyFavoured in 18 of 24
After TNF-inhibitor failureFavoured in 6 of 24
Pregnancy planning
Not shown. Compared fewer than 8 times in this profile, or not at all.
Cardiovascular & thromboembolic riskFavoured in 8 of 16
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Tocilizumab head to head →Follow Tocilizumab across the patient profiles →
AI Visibility
Is it named at all?
23%
414 of 1,816 answers · rank 5 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
8%
116 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
11 · 8%Favoured
44 · 31%Comparator favoured
144 judgments against 13 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapy
Not shown. Compared fewer than 8 times in this profile, or not at all.
After TNF-inhibitor failureFavoured in 0 of 32
Pregnancy planning
Not shown. Compared fewer than 8 times in this profile, or not at all.
Cardiovascular & thromboembolic riskFavoured in 0 of 20
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Tofacitinib head to head →Follow Tofacitinib across the patient profiles →
AI Visibility
Is it named at all?
21%
381 of 1,816 answers · rank 6 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
6%
87 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
16 · 7%Favoured
83 · 38%Comparator favoured
220 judgments against 12 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapyFavoured in 1 of 12
After TNF-inhibitor failure
Not shown. Compared fewer than 8 times in this profile, or not at all.
Pregnancy planningFavoured in 0 of 8Small base
Cardiovascular & thromboembolic riskFavoured in 4 of 12
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Adalimumab head to head →Follow Adalimumab across the patient profiles →
AI Visibility
Is it named at all?
20%
363 of 1,816 answers · rank 7 of 25 scored therapies
Named first
When a scored therapy is named, is it this one first?
3%
39 of 1,496 answers naming any scored therapy. Order of mention, not preference.
Head-to-head verdicts
When an answer compares it with another therapy, which does it favour?
65 · 41%Favoured
11 · 7%Comparator favoured
160 judgments against 12 opponents, including pairs compared fewer than 8 times.
Patient profiles
Does the verdict change with the patient?
MTX intolerance / monotherapy
Not shown. Compared fewer than 8 times in this profile, or not at all.
After TNF-inhibitor failureFavoured in 16 of 32
Pregnancy planning
Not shown. Compared fewer than 8 times in this profile, or not at all.
Cardiovascular & thromboembolic riskFavoured in 0 of 24
Each profile has its own questions and comparators. Not a controlled comparison.
Compare Upadacitinib head to head →Follow Upadacitinib across the patient profiles →
Your portfolio
Go beyond visibility. See how AI compares your therapy with its competitors, on which questions and for which patients.
About this benchmark
AI representation analysis of 1,816 answers to US rheumatologist-style open questions and 960 answers to comparative questions from four web-enabled AI systems (ChatGPT, Claude, Gemini, Perplexity), collected 30 October 2026 with a US rheumatologist persona. Visibility = share of answers naming a therapy. Neither visibility, order of mention nor head-to-head verdicts indicate clinical efficacy, safety, preference or recommendation.
Pre-ACR baseline (v1.1 corrected analytical view, 30 Oct 2026). Re-measured after ACR Convergence 2026.
One pre-ACR Convergence 2026 baseline. ChatGPT, Claude, Gemini and Perplexity, each with web search, answered the same questions once each on 30 October 2026, with a US rheumatologist persona applied. The answers are AI answers.
AI Visibility — the share of all 1,816 answers to open questions that name a therapy, by its molecule name or any in-scope brand, matched word for word against a frozen list of 25 molecules and 43 brands. Any mention counts, including cautionary or historical mentions. Names that appear only in a link are not counted.
Named first — the first scored therapy named in an answer, over the 1,496 answers that name at least one. Order of mention, not preference. It is reported for the four AI systems combined, for the ten therapies named first most often.
Head-to-head verdicts — an AI judge (gpt-4.1-mini) read each of the 960 answers to comparative questions and scored each therapy the question names as preferred, equal, inferior or not compared. A therapy is favoured only when it was preferred, its comparator inferior, and the judge noted no condition. Conditional verdicts — where the answer makes the choice depend on the patient or the clinical context — are shown separately and never counted as favoured. Other covers answers where only one therapy was scored, no clear preference was expressed, or a therapy was not mentioned. Eight complex questions were reviewed by an analyst instead and are not in the figures.
Patient profiles — a comparative answer belongs to a profile when its question is framed around that patient. Profiles overlap, so their figures are never added together. Four profiles are published, and within each only the therapies compared at least 8 times.
Small bases — the base is stated wherever a figure rests on fewer than 30 answers, and a figure resting on fewer than 10 is marked Small base. Many question-type rows of a single pair rest on 4 answers: one question, answered once by each AI system.
What the explorer offers — the head-to-head explorer offers the 42 pairs compared at least 8 times. A therapy's total head-to-head verdicts, and the JAK question-type view, also count pairs compared fewer times, so they are larger than the sum of the pairs the explorer shows.
Not efficacy, safety or clinical preference, and not a recommendation. A verdict describes what one AI answer favoured for one question; pooled verdicts do not establish clinical superiority, and no net or overall score is computed. No AI system is ranked or scored: differences between systems are mostly how many therapies an answer names. Therapies outside the scored list are not ranked and are not zero. One collection date, before ACR Convergence 2026; web-search answers change over time, and this benchmark will be re-measured after ACR.