Can AI Find Your Medical Society

Ask an AI tool about your society by name and it will usually give a reasonable description and maybe a link. Association leaders run that test, get a good result, and conclude the AI question is handled. The questions that decide whether a prospective member ever finds you are the ones with your name left out, like which societies a surgeon in this specialty should consider joining, who publishes the guideline for this condition, what grants exist for early-career researchers in this field. On those, the results get uneven. The society sometimes goes unmentioned entirely, and when it does appear the citation often points at a journal or a university instead, or a larger neighboring organization takes the whole answer. Being retrievable by name and being selected as the authority are separate problems, and most of the advice being sold to associations addresses only the first.

The audience has moved faster than most societies realize, and the research on how AI tools choose sources says societies should be doing well here. AI engines answering health questions lean heavily on institutional sources, and professional associations are among the types they favor most.

The evidence in this piece is drawn from primary sources throughout, including the American Medical Association’s 2026 physician survey on augmented intelligence, Pew Research Center’s study of Google users and AI summaries, the AP-NORC Center’s survey of AI adoption, and two 2026 analyses of how AI tools cite health sources, one studying ChatGPT and one studying Claude. Where a figure comes from a company’s own reporting rather than independent research, the text says so. Where we describe how AI tools handle particular kinds of questions, treat it as a snapshot, because results vary by platform, date, and session, and any society should test its own questions rather than assume.

Your members already work inside AI

Start with the people who already pay dues. A survey of 1,692 US physicians conducted early this year found 81 percent now use AI in their professional practice, up from 66 percent in 2024 and 38 percent in 2023. The fastest-growing use is the one closest to what a society does, summarizing medical research and standards of care, which 39 percent now use AI for, up 26 percentage points in a single year.

Much of that use runs through one tool. OpenEvidence, which is free to verified clinicians, told a news investigation it is actively used by roughly 650,000 US physicians, about 65 percent of the profession, and that it handled almost 27 million clinical consultations in April 2026 alone. Those are the company’s own figures rather than audited ones, though NEJM Group and the JAMA Network both signed content licensing agreements with it in February and June of 2025, which says something about how seriously the publishers take it.

Those agreements connect the medical literature layer directly to the tools physicians use at the point of care, and physicians report trusting these tools partly because the answers carry citations. Your members encounter cited clinical knowledge through an AI intermediary many times a day, so whether your society’s guidelines and consensus statements sit inside that citation layer describes how your members work now rather than how they might work later.

The people you want to recruit search differently now

The recruitment side has moved just as far. A probability-panel survey of 1,437 US adults in July 2025 found 60 percent had used AI to search for information, the most common AI use it measured, and among adults under 30 it was 74 percent. Those under-30s are your residents, fellows, and early-career members, which is the cohort where first-year retention already runs weakest, and a growing share of them will meet your society for the first time inside an AI answer rather than on your homepage.

The tools have the scale to matter. ChatGPT reached 900 million weekly users in February 2026 by its maker’s own count, double the year before. A growth agency’s analysis put AI assistant sessions at the equivalent of 56 percent of global search engine volume, and its most conservative cut, counting only prompts that directly compete with search, still came to 28 percent worldwide. Where AI answers appear inside traditional search they absorb the clicks, and browsing data covering 68,879 real Google searches found users clicked a result 8 percent of the time when an AI summary was present, against 15 percent when it wasn’t.

This lands on a problem societies already knew they had. In the industry’s own member research, the top reason nonmembers give for never joining is that they didn’t know a relevant association existed, ahead of cost, and a third of association executives say the same thing about their prospects. Awareness was already the top acquisition barrier when discovery meant Google, and discovery is now moving to tools that return one synthesized answer instead of ten links, which puts the awareness problem in the hands of whatever decides which organizations get selected and credited.

One aside for readers on the health-tech side, since the same shift is arriving in B2B. A survey of 1,076 software buyers in April 2026 found 51 percent now begin research with an AI chatbot more often than with Google, up from 29 percent a year earlier, and a third had bought from a vendor they had never heard of before the AI surfaced it.

What the citation research says about societies

When researchers look at which sources AI tools actually cite for health questions, the results should encourage a society executive.

A 2026 analysis of ChatGPT’s responses to consumer health questions found more than 75 percent of cited sources were established institutional ones, with the top ten organizations accounting for over half of all citations. A companion study of Claude coded 10,038 citations across 3,075 health questions and found institutional sources made up nearly 98 percent of the total, broken down as medical institutions at 36.5 percent, government sources at 31.6 percent, and professional associations at 28.4 percent. A separate industry analysis of 5,472 citations across four major AI tools found the single most-cited domain for health information was PubMed Central, the open repository of peer-reviewed medical literature.

Those numbers describe a professional society. AI engines answering medical questions select institutional, specialized, authoritative sources, and more than a quarter of one major tool’s health citations already go to professional associations. So the open question is why, on the questions a society is the actual authority on, the citation so often lands somewhere else.

Why societies lose unbranded questions

These losses happen at the moment of selection rather than at indexing, and a few mechanisms account for most of them.

The publisher gets the credit. A society writes the guideline, runs the consensus process, or funds the research, and the most visible version of that work ends up on a publisher’s platform or in a database of record. An AI tool cites where it read the content rather than who created it, so the substance of the answer is the society’s work while the credit reads as someone else’s.

The expertise is scattered. The answer to a question the society owns exists, but in pieces: a meeting abstract, a PDF archive, a program page on a subdomain, the full text on a publisher site. Nothing states the answer plainly in one place, in the words a physician would use to ask it, so the tool builds its answer from whoever does have a single clear page, which is often a hospital system, a government site, or a larger organization.

The larger neighbor takes the category. Ask which society someone in a specialty should join and these tools tend to favor the organization with the clearest public membership and resource pages, which is usually the biggest one. A smaller society can be entirely findable by name and still lose every version of the joining question to the organization next door.

The site speaks in institutional language. Societies name programs after donors, committees, and history, and organize pages around the org chart. Physicians ask about grants for early-career researchers, registries for a procedure, which meeting is worth attending. When the page’s language doesn’t match the question’s language, the connection gets left for the machine to infer, and often it doesn’t.

Some of it can’t be read at all. Content behind member logins and inside PDFs stays invisible to the systems doing the selecting, and the citation research is consistent that what gets cited is open, current, and structured. Readability is the floor here rather than the strategy.

One caution against the easy scapegoat: A legacy website platform doesn’t automatically cause any of this, and the available evidence doesn’t support blaming the CMS. A modern platform makes the fixes easier to implement, though the durable issue is whether the society’s expertise is clearly stated and connected to the questions its audience actually asks.

Not every question is yours to win

Manufacturing visibility for questions another organization legitimately owns wastes effort, and it reads as inflated to the exact audience a society is trying to earn. So the work starts with sorting.

Questions your society must own are the ones where it is the definitive authority: your guidelines, your registry, your grants, your meetings. An AI answer to one of these that leaves you out is the emergency. Questions where you should appear are the broader specialty questions where you’re one of several legitimate authorities and absence is a real loss. Questions where you could reasonably appear are adjacent topics, where inclusion is a bonus rather than a right. And questions another organization owns are the national database a larger society runs or the formal guideline a different body publishes. Conceding those plainly is part of what makes your claim to the first category credible.

Most societies have never made this list, so nobody at the organization can currently say which AI answers represent a problem and which represent the world working correctly.

The twenty-minute AI visibility test

You don’t need a consultant to find out where you stand, and you shouldn’t take anyone’s word for how these tools treat your society.

Write down five to eight unbranded questions covering what your society actually does, phrased the way a clinician would ask them, with your name left out. Cover joining, guidelines, funding, registries and quality programs, and meetings or education.

Ask each one, in incognito, in ChatGPT, Google’s AI Mode, Claude, and Perplexity. For every answer, record whether your society is named, whether your own site is cited or the credit goes elsewhere, and who gets selected in your place.

Then read the results against the four categories above. A must-own question answered without mention of your society writes your priority list for you. A should-appear question lost to a larger neighbor tells you which first-party page to build next. A question another organization owns, answered with that organization, goes in the file marked working as intended.

Run the test twice a few days apart before concluding anything, and date your notes. These answers move between tools and sessions, and the pattern is what you’re after rather than any single response.

What actually works for AI visibility

The practical work follows from the failure patterns, and one popular tactic deserves clearing away first.

Skip llms.txt. A proposed standard by that name gets sold widely as an AI visibility fix. No major AI engine honors it, Google’s search team has said so directly and compared it to the long-discredited keywords meta tag, and an analysis of 137,000 sites found 97 percent of llms.txt files had received zero visits from anything.

Check crawler access, then move on. AI systems can only select what their crawlers can read, so confirm your robots.txt isn’t blocking GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, or Google-Extended. Some platforms block them by default without telling anyone. This takes five minutes and it’s necessary, but it’s hygiene rather than strategy.

Build one first-party page per question you must own. One open public page for each, answering the question directly, with the answer first, plain headings, a FAQ section, current dates, and proper schema markup. State explicitly what the society publishes, funds, administers, and operates, using the language of the question. A grant page that says who and what it funds, a guidelines page that presents the recommendations instead of linking out to a journal, or a membership page that answers why and for whom.

Claim your attribution. When your strongest intellectual property lives with a publisher, build the first-party companion, which is the society’s own summary page for each guideline and consensus statement, naming the society as author and linking to the full text. This duplicates nothing, and it gives the selecting systems a first-party source to credit. The depth, the tools, and the community stay behind the login while the authoritative summary lives in public, which is the trade every open-access abstract already makes.

Organize around questions rather than departments. The information architecture that helps is the one that mirrors how physicians ask: joining, guidelines, funding, registries, meetings, careers. If a resource matters, its page should say what it is in the first sentence, in words someone outside the building would use.

The window

The field’s own numbers say how much time this buys a society that moves now. In the association industry’s 2026 benchmarking data, 23 percent of associations use AI in their membership marketing at all, the share using AI for search optimization fell from 33 to 23 percent year over year, and 9 percent of the associations using AI feel ready for what’s coming. The trade press has started telling associations to structure their pages so AI tools can surface their expertise, and consultants at the field’s technology conferences now acknowledge that content behind member paywalls can’t be seen or learned from. What doesn’t exist anywhere we could find is a medical or professional society that has published its own strategy for this, or any rigorous measurement of how society content actually fares in AI answers.

Your members aren’t waiting for the field to catch up. Eighty-one percent of physicians are already using these tools, your prospective members are the demographic adopting AI search fastest, and the answers to your specialty’s questions are being assembled right now out of whichever organizations state their authority clearly enough to be selected.

Frequently asked questions

Is my medical society invisible to AI tools? Probably not, and that’s the trap. Most societies are findable when asked about by name. The gap shows up on unbranded questions, like which societies to join, who publishes a guideline, or what grants exist, where AI tools often fail to select the society, cite a publisher or university instead, or prefer a larger organization. Being findable and being selected are separate problems.

Do AI tools like ChatGPT cite medical societies and professional associations? Yes, heavily, when the content is selectable. Research on AI health citations found professional associations account for 28.4 percent of one major tool’s health citations, and more than 75 percent of ChatGPT’s cited health sources are established institutions. The common failures are attribution going to publishers, fragmented content with no clear first-party page, and program language that doesn’t match how physicians ask.

Why does my association show up when someone names it but not for broader questions? Because being retrievable and being selected work differently. AI can describe almost any organization by name. At unbranded questions, tools select whichever organization has the clearest first-party answer, and societies whose expertise is scattered across PDFs, publisher sites, and internal-language program pages lose those selections even when the expertise is theirs.

How do I find out how AI tools treat my association? Test it in twenty minutes. Pose five to eight unbranded questions across joining, guidelines, funding, registries, and meetings in ChatGPT, Google’s AI Mode, Claude, and Perplexity, and record whether you’re named, whether your site gets the citation, and who is selected instead. Sort the results by whether you must own, should appear on, or don’t honestly own each question, and date your notes, since results shift over time.

What is generative engine optimization for associations? Making an organization’s expertise selectable and correctly attributed by AI systems: first-party answer pages for the questions the organization owns, written in the language its audience uses, clear public statements of what it publishes, funds, and operates, consolidated resource hubs instead of scattered PDFs and subdomains, crawler access, and structured markup. The work is closer to institutional clarity than to a technical trick.

Does llms.txt improve AI visibility? Not based on current evidence. No major AI engine honors the proposed standard, Google has said it does not use it, and a large-scale analysis found 97 percent of llms.txt files received no visits. Crawler access and clear first-party content are what the citation research supports.

Why does AI search matter for membership recruitment? Because awareness is the top reason prospects never join, and discovery is moving to AI. Sixty percent of US adults have used AI to search, 74 percent of those under 30, and clicks on traditional results roughly halve when an AI summary answers first. A young physician’s first encounter with your society may now be an AI answer, and whether you appear in it is being decided by the clarity of your public content today.