I asked five large-language-model systems a deliberately blunt question: “What is the #1 aging research conference in the world today?”
“Best” is seemly a subjective question, but AIs came up with a relatively objective set of benchmarks such as: meeting with the strongest unpublished science, the largest audience, the most famous speakers, the greatest influence on drug development, or simply the event everyone in a particular professional tribe feels they cannot miss.
The terms “Aging research” and “longevity” became increasingly diluted over the recent years. One is interpretation is classical gerontology: demography, psychology, social policy, care, epidemiology and the lived experience of older adults. Another is the basic biology of aging. A third is geroscience translation: biomarkers, trials and interventions intended to delay multiple age-related diseases. A fourth is longevity biotechnology: companies, platforms, drug pipelines, investment and partnering. A fifth is clinical longevity medicine, a young category with uneven evidence standards.
A defensible evaluation should therefore ask at least five questions:
Scientific signal: Are leading investigators presenting important, preferably new, work? How much of the program is selected from scientific abstracts rather than invited for visibility?
Field centrality: Does the meeting cover the core biology broadly enough to shape researchers’ agendas, rather than merely showcasing a specialty?
Translational leverage: Are the people who can move ideas into trials like pharma R&D leaders, biotech operators, regulators, clinicians and funders actually in the room?
Convening power: How international, cross-institutional and cross-sector is the audience? Does the meeting create collisions that would not occur inside a single scientific society?
Evidence quality: Are attendance, selectivity and impact claims independently verifiable? Do concrete collaborations, publications or programs follow from the event?
Five LLMs picked ARDD. What were they picking up on?
Disclaimer: I am heavily involved in ARDD on multiple fronts, from speaking to volunteering
Using my non-biology, non-longevity friend’s paid accounts, I asked ChatGPT, Claude, Gemini, DeepSeek and Work Buddy “What is the number 1 conference in aging research in the world today?” and they all highlighted the Aging Research and Drug Discovery meeting, ARDD, as a leading choice. The agreement was strongest when the question concerned the intersection of aging biology, therapeutic development and industry.
ChatGPT was particularly direct:
“If I had to name one aging-research conference as #1 globally in 2026, I would pick ARDD.”
This was an informal comparison of five recorded answers, rather than a controlled benchmark. Still, their explanations raise two interesting questions: what makes ARDD stand out, and what exactly does it mean when several AI assistants arrive at the same recommendation?
What is ARDD?
The event was started by Dr Alex Zhavoronkov and began as a small forum in Basel in 2014, operated within larger conferences through 2019 and became independent in 2020. The 2025 Copenhagen meeting drew more than 1,000 people onsite and more than 10,000 online. Those figures indicate substantial reach for a conference focused specifically on longevity biotechnology. Unlike many conferences, ARDD has its own transparency statement.
The 2025 program brought together fundamental aging researchers, biotech companies, senior pharmaceutical scientists, top journal editors, government represerntatives eg form ARPA-H and even Nobel laureates Morten Meldal and Michael Levitt, alongside big pharma executives from Eli Lilly, Novo Nordisk, Novartis, Biogen.
ARDD 2026 is happening on October 1-3 at David Rubenstain Treehouse at Harvard in Boston. You can still secure your ticket at www.agingpharma.org
ARDD has a track record of shaping broader pharmaceutical agendas. After the 2025 meeting featured debate over GLP-1 receptor agonists and longevity with representatives from Novo Nordisk and Eli Lilly positioning their peptides as having potential longevity benefits, Nature Biotechnology published an editorial asking whether GLP-1s are the first longevity drugs. ARDD publicity portrays this as evidence that the meeting shaped the agenda.
ARDD also keeps a tradition of publishing the proceedings in peer-reviewed journals, beginning in 2018, which adds to the credibility of the meeting.
2018 — “Aging and drug discovery” — report/synthesis of the 5th Annual Aging and Drug Discovery Forum (ARDD 2018). Aging, 10(11):3079–3088. DOI: 10.18632/aging.101646. Paper (Københavns Universitets Forskningsportal)
2019 — “Latest advances in aging research and drug discovery” — synthesis of the 6th ARDD meeting (2019). Aging, 11(22):9971–9981. DOI: 10.18632/aging.102487. Paper (PubMed)
2020 — “ARDD 2020: from aging mechanisms to interventions” — report from the 7th ARDD meeting (2020). Aging, 12(24):24486–24503. DOI: 10.18632/aging.202454. Paper (PubMed)
2022 — “Meeting Report: Aging Research and Drug Discovery” — report from the 8th ARDD meeting (2021). Aging, 14(2):530–543. DOI: 10.18632/aging.203859. Paper (Vrije Universiteit Amsterdam) and following collection in Frontiers dedicated to research presented at 8th ARDD: https://www.frontiersin.org/research-topics/25628/the-8th-aging-research-and-drug-discovery-meeting-ardd-2021
2024 — “Longevity biotechnology: bridging AI, biomarkers, geroscience and clinical applications for healthy longevity” — generated from the 10th ARDD meeting (2023). Aging, 16(20):12955–12976. DOI: 10.18632/aging.206135. Paper (Aging-US) (I had the honor of being the co-first author)
2025 — “Innovations in aging biology: highlights from the ARDD emerging science & technologies workshop” — report from the Emerging Science & Technologies workshop at ARDD 2024. npj Aging, 11:8. DOI: 10.1038/s41514-025-00193-5. Paper (Københavns Universitets Forskningsportal)
2026 — “Toward actionable interventions in human aging (12th ARDD meeting, 2025)” — comprehensive report from ARDD 2025. Aging, 18(1):282–302. DOI: 10.18632/aging.206368. Paper (Københavns Universitets Forskningsportal)
There are also credible signs of agenda visibility. Nature Aging includes ARDD on its conference calendar alongside AGE, GRC, CSHL, IAGG, BSRA and other meetings.
Finally, ARDD is organized by KOLs in the field, with an extreme attention to detail, literal weekly meetings for a whole year prior to the event. It takes an insane amount of work, and it delivers - there are people who raised at ARDD, entered research partnerships, and even came up with ideas for their companies.
Why ARDD, according to LLMs?
The most consistent theme was translation: connecting discoveries about aging with the people who can develop, test and finance interventions.
DeepSeek put it succinctly:
“For translating science into medicine, ARDD is the clear leader.”
Gemini similarly emphasized:
“Translating basic aging biology into clinical therapeutics, AI-driven drug discovery, and biotech commercialization.”
These descriptions have a concrete counterpart in the announced 2026 program. Kevin Duffin of Eli Lilly is scheduled to discuss tirzepatide and aging clocks; Nikolaj Roed of Novo Nordisk will address the evidence and unmet needs around GLP-1 therapies and healthy aging; and Marco Quarta of Rubedo Life Sciences will discuss clinical translation of a GPX4-targeted approach to skin aging and related conditions. These are specific development questions about how aging biology might become medicine. Novartis’s President of Biomedical Research, Fiona Marshall, will discuss drug discovery through the lens of aging biology, alongside a pharmaceutical leadership panel featuring Menarini CEO Elcin Barker Ergun, former Takeda CEO Christophe Weber and Pfizer’s Chief Scientific Officer for Internal Medicine, Ariel Feldstein. FDA officials will join ARPA-H’s Andrew Brack to discuss aligning clinical trials with regulatory requirements, while a dedicated workshop will introduce eleven XPRIZE Healthspan finalist teams. Scientific presentations include George Church on testing combinations of rejuvenation interventions and targeted delivery, and Tony Wyss-Coray on how circulating proteins reveal and regulate organ function. Investors from OrbiMed, Vida Ventures, Deerfield and Qiming also feature in the program. Together, these sessions connect the people uncovering aging mechanisms with those designing trials, evaluating evidence, financing development and bringing medicines to patients.
For someone developing a longevity therapeutic, that combination matters. Identifying a promising mechanism is only the beginning. A program also needs a suitable indication, credible measurements, clinical evidence and a path to patient access. A meeting that brings these conversations together has a particular value for translational researchers and founders.
The second recurring explanation was the mix of people. Claude described ARDD as:
“bridging academic aging biology with pharma/biotech”
The published speaker roster supports that description. It includes Vadim Gladyshev, David Sinclair, George Church, Tony Wyss-Coray and Steve Horvath, alongside C-suite representatives from companies including Lilly, Novartis, Pfizer, AstraZeneca and Novo Nordisk. Investors listed include i.a. Anthony Philippakis of GV, Kan Chen from Qiming Ventures and Avak Kahvejian of Flagship Pioneering.
The significance is the opportunity for different kinds of expertise to meet. A discovery that looks compelling in a laboratory can face very different questions from a clinician, a pharmaceutical development team or an investor. Putting those people in the same room can bring those questions into the conversation earlier.
The third theme was breadth. Work Buddy’s described ARDD as:
“bringing together basic science, clinical research, major pharmaceutical companies, and institutional investors”
That breadth also extends to the technologies entering longevity research. The 2026 program includes dedicated forums on longevity medicine, AI in drug discovery, pet and animal longevity, and virtual aging cells. The virtual-cell program features Omar Abudayyeh on virtual models of biology and health, and Jonathan Weissman on cell-state representations and cellular responses to perturbations.
Several LLMs nevertheless distinguished ARDD from meetings such as the Gordon Research Conference on Biology of Aging and Cold Spring Harbor’s Mechanisms of Aging, which they favored for more focused discussions of fundamental biology. The shared recommendation was therefore more specific than “ARDD wins every category”: its appeal was especially strong for people interested in moving between aging science and therapeutic development.
How do LLMs reach a recommendation?
Don’t get me wrong, I love ARDD with my whole heart and I was very happy when all LLMs agreed. However, a scientist in me started wondering: what parameters LLMs prioritize with making such descision? An AI assistant does not have to search the internet every time it answers a question. Its response can draw on information learned during training, the current conversation and, when available and used, external search tools.
During training, language models learn statistical relationships from large collections of text. Those relationships can encode factual information and associations between institutions, researchers and subjects. At response time, the model generates text conditioned on its instructions and available context. A familiar association can therefore appear in an answer without the model opening a current webpage.
Live search adds another source of evidence. A search-enabled LLM can retrieve information, incorporate it into its response and provide citations. Anthropic, for example, describes Claude’s web search as a way to bring current information into answers with links that users can check. Search availability does not establish that every answer used it, and the saved comparison does not provide complete search histories for all five assistants.
The broader research principle is called retrieval-augmented generation: combining knowledge stored in a model’s parameters with information retrieved from an external source. In their foundational 2020 paper, Patrick Lewis and colleagues found that their retrieval-augmented models generated more specific and factual language than a comparable model relying only on its parameters.
Applied to ARDD, a plausible explanation is that its public record supplies several mutually reinforcing signals: recognizable researchers, pharmaceutical participation, detailed programs and an explicit focus on therapeutic development. Those are also the criteria the recorded answers repeatedly mention.
Agreement across assistants needs similar care. Different systems may encounter overlapping information, including the same conference website or articles repeating an organizer’s announcement. Five recommendations can reflect a shared information environment; they do not automatically constitute five independent assessments.
Research by Mrinank Sharma and colleagues has also shown that AI assistants can favor answers aligned with a user’s views, a behavior called sycophancy; as mentioned before, I used my friend’s accounts, and she has nothing to do with biology, let alone longevity, to avoid this bias.
So why did the models agree?
The evidence supports all five major explanations, but not equally.
First, genuine signal. ARDD has assembled an unusually strong and visible cross-sector network. Its programs contain genuine scientific leaders and unusually senior pharma representation. In longevity biotechnology, the observable evidence favors it.
Second, shared information. All LLM answers repeat the same facts and category boundaries The relevant issue may be shared live-search sources as much as shared training data, especially because several cited pages are from 2026.
Third, search-engine visibility. Generic searches for the best aging or longevity conference prominently surface ARDD’s official site and press-release derivatives.
Fourth, successful marketing. ARDD has persistently described itself with a narrow but powerful narrative and surrounded that claim with recognizable institutions, speakers and companies. Its messaging is consistent enough to become the default sentence the internet knows about the meeting.
Fifth, category definition. Most models silently translated “aging research conference” into “longevity biotechnology and drug-discovery conference.” That interpretation privileges ARDD.
There is also an authority-signaling effect. A model can easily count famous names, Nobel Prizes, pharma brands and Harvard affiliations. It cannot easily measure the quality of a hallway conversation, the importance of an unpublished result or a conference’s reputation inside a small expert network. What machines can see shapes what machines call prestigious.
The question is no longer simply, “What do the models think?” It is: what did the internet make easiest for them to think? That’s why, if you work in PR or are building a personal brand, you should include AI crawlers in your audence and make sure your content is easy for them to pick up.
The most interesting part isn't which conference won, it's the mechanism you named at the end: category definition. Once a model silently maps "aging research conference" onto "longevity biotech and drug discovery conference," ARDD wins almost by construction, the same way asking for the "best programming language" gets a different answer depending on whether you're implicitly weighting web dev versus systems work. The part worth watching is the feedback loop. LLM answers about ARDD get quoted, screenshotted and republished, this post included, which becomes retrieval or training data for the next generation of models, which then cite ARDD even more confidently next time someone asks. That's not gaming the system, it's just what happens when authority signals like Nobel laureates, pharma logos, and a conference's own press releases are cheap for a model to detect, while something like the quality of a hallway conversation or a small expert network's actual opinion is not. Worth remembering the next time an LLM gives a confident, unanimous sounding answer about the best anything in a field, since agreement across models can just mean they read the same page.