Everyone in Pharma Is Optimizing for AI. Almost Nobody Planned for the Customer.
GEO, AEO, AIEO, LLMO. The letters keep multiplying. Meanwhile your doctors and patients are already living inside these tools, and how they use them decides everything the acronym debate is busy ignoring.
The short version
- The most-used clinical AI tool among US physicians is not ChatGPT. It is OpenEvidence, used by roughly two thirds of them, evidence-grounded, and funded by pharmaceutical advertising.
- Patients are already here too. By OpenAI's own account, around 40 million people a day bring health questions to ChatGPT.
- HCPs and patients use different engines, with different indexes and different rules. Optimising for one does almost nothing for the other.
- For pharma the hard part is not visibility, it is the regulatory wall. You cannot be the answer to "best treatment," and the model paraphrases past the safety information you are legally required to carry.
- Planning for it is a technical stack, not an acronym: be crawlable, sit in the authoritative band of the right index, get validated by sources others trust, structure content for extraction, and accept you cannot control the framing.
Sit in any life sciences meeting about AI right now and you can play a quiet game. Count the acronyms. SEO had a child called GEO. GEO has a cousin called AEO. AEO gets mistaken for AIEO, which shares a flat with AIO and LLMO. The industry has turned "plan your content for AI" into a contest over which three letters win, and the honest part is that most of the people saying them cannot define them. There is no agreed taxonomy, and fewer than a third of search influencers use the same term consistently inside a single article they wrote themselves.
While that argument runs, a more important one never starts. Which AI. Used by whom. To ask what. Inside which rules.
Because here is what the acronym debate keeps missing: your customers did not wait for it.
The doctor already left the room, and she is on OpenEvidence
The most-used clinical AI tool among US physicians is not ChatGPT. It is OpenEvidence, a purpose-built answer engine grounded in the literature doctors already trust, NEJM, JAMA, the NCCN guidelines. By spring 2026 it was being used by roughly two thirds of US physicians across tens of millions of consultations a month. It is free. It is funded by pharmaceutical advertising. Read that twice, because it means there is already a paid AI surface sitting between your brand and the prescriber, and most commercial teams have never once discussed it.
Step back and the trend is just as steep. The AMA's 2026 survey found 81% of physicians now use AI in practice, more than double the rate three years earlier, with literature search and documentation leading. Outside the US the picture is lower and more fragmented. Around a quarter of UK GPs report using these tools clinically, and the evidence layer they trust is different again, NICE and the EMA label rather than the US journals. Same behaviour, different plumbing by geography.
The patient is already typing into ChatGPT
On the consumer side the numbers are larger and stranger. By OpenAI's own account, around 40 million people a day bring health questions to ChatGPT, and more than 5% of everything it handles is health-related. Three in five US adults say they have used AI for a health question in the past three months. To check a symptom. To understand a term their clinician used. To weigh a treatment option. To make sense of an insurance letter. The single biggest area is not the one most brand teams would guess. It is mental health and companionship, with roughly one in eight young people now using a chatbot for emotional support. Oncology is the most-studied disease area, where patients use these tools to ask the frightened questions they leave out of the actual appointment.
They use it off-hours. Between visits. When the clinic is closed and the human is not available. None of them is refreshing a slide to see whether we landed on GEO or AEO.
Why this is harder for life sciences than for anyone else
Most industries can simply chase visibility. Pharma cannot, and this is the part the roundtables skip.
There are two doors, not one. The HCP door is the evidence layer. You earn presence by being inside the cited literature and guidelines an engine like OpenEvidence draws from. The patient door is general-purpose, mostly ChatGPT and Google's Gemini. They are different engines, different indexes, different rules, and optimising for one does almost nothing for the other.
Then there is the wall nobody mentions over coffee. You are not allowed to be the answer to "best treatment." That is not a tactic you forgot, it is a regulation. And even when an engine cites you, it does not reproduce your page. It paraphrases. It can strip the safety information you are legally required to carry, soften the fair balance, and occasionally invent an indication you never had. You do not publish that answer. The machine does, in its own words, and your medical reviewers were never in the room. Models also frame. Ask one why a clinician might prefer chemotherapy over immunotherapy and it will often argue the case rather than weigh both sides. For a regulated brand, that is not a curiosity. It is exposure.
So "rank for AI" is the wrong instinct. The right one is narrower and more useful. Know the surface, the question, and the rule you cannot break.
How to actually plan for the word salad
Here is the part that turns the vibe back into work. None of it is a single trick. It is a stack, in order of how much it matters.
Be readable by the machines first. The most common and most invisible failure is a site that quietly blocks AI crawlers, often through a default Cloudflare or plugin setting nobody chose on purpose. If OpenAI's search crawler, Perplexity's, and the rest cannot reach your pages, you cannot be cited at all. There are three kinds of bot to think about: the ones that train models, the ones that build the search index, and the ones that fetch a page live when a user asks. Allow the retrieval ones. And learn the traps, because blocking Google's training opt-out does nothing to your presence in Gemini, which still runs off the ordinary Google crawler.
Position still does the heavy lifting, but the target moved. The engines ground their answers on search indexes, so where you sit in search still drives whether you get pulled into a response. What changed in 2026 is that the citation no longer comes from the top of page one. One large March 2026 analysis by Ahrefs found under 38% of AI-cited URLs ranked in the top ten organic results, down from around three quarters the previous summer. The new sweet spot is the broad authoritative band: trusted enough to be pulled, not necessarily winning the old race.
Be validated by others, not just by yourself. The original Princeton research on generative engine optimization found that citing credible sources and adding real statistics are among the highest-impact things you can do. In life sciences that maps cleanly onto the things that already carry weight: peer-reviewed publications, society guidelines, the evidence base the clinical engines pull from. Your own beautifully approved PDF is the weakest card in the deck.
Then the unglamorous mechanics. Structure content so a machine can lift a clean passage. Clear headings. A definition near the top. Real data. Recent dates, because freshness is brutal and the large majority of citations come from content updated within the last year. And pick your engine by audience, because Gemini reads Google, Copilot reads Bing, Perplexity and ChatGPT read their own indexes, and Grok reads the social graph.
Last, the legal weather, because it decides what the engines are even allowed to touch. Publishers are suing and licensing in roughly equal measure, and the defence the AI search companies keep returning to is that you cannot copyright facts. That tells you exactly what survives. Your data and your numbers can be lifted and resynthesised. Your framing cannot be controlled. The content that wins is the authoritative, properly sourced kind that still holds up when something you do not own pulls it apart and reassembles it.
The customer was there the whole time
The acronym is the easy part. It is a costume the industry keeps trying on. The body underneath is the doctor already working inside OpenEvidence, and the patient already asking ChatGPT something at two in the morning that they were too embarrassed to ask you.
Plan for them. Know which engine they use, what they ask it, and which rule you are not allowed to break when it answers. That is not an SEO project with a fresh vowel. It is customer strategy, and it is the work almost nobody in the room is actually doing.
FAQ
What is GEO, and how is it different from SEO, AEO and AIEO? GEO (generative engine optimization) is the practice of getting your content cited inside AI-generated answers rather than ranked in a list of links. AEO (answer engine optimization), AIEO (AI engine optimization) and LLMO describe the same underlying work under different names. There is no industry consensus on the terminology, and in practice they overlap almost completely.
Which AI tools do doctors use most? For day-to-day clinical questions, US physicians overwhelmingly use OpenEvidence, a purpose-built engine grounded in medical literature. General-purpose ChatGPT remains the most-used tool for everything else, and overall physician AI use reached 81% in 2026. Adoption outside the US is lower, and the evidence base clinicians trust differs by country.
How do patients use ChatGPT and other LLMs for health? Patients use them to check symptoms, understand medical terms and instructions, weigh treatment options, and make sense of insurance, with around 40 million people a day asking ChatGPT health questions. The single largest area is mental and emotional support. Most of this happens off-hours and between clinical visits.
Can a pharma company optimise its content to appear in AI answers? Yes, but inside tighter limits than other industries. You cannot promote a brand as the "best" treatment, and you cannot control how a model paraphrases your label or whether it carries the required safety information. The realistic, compliant target is unbranded disease education and the authoritative evidence base that clinical engines cite.
How do you get cited by ChatGPT, Gemini and Perplexity? Make sure AI crawlers can read your pages, sit in the authoritative band of the relevant search index, earn citations from third-party sources, and structure content so a passage can be lifted cleanly. Keep it current, because most AI citations come from recently updated content. Optimise per engine, since each grounds on a different index.
Do HCPs and patients use the same AI tools? No. Clinicians lean on evidence-grounded clinical engines like OpenEvidence, while patients use general-purpose tools like ChatGPT and Gemini. These run on different indexes with different rules, so a single optimisation approach will not reach both audiences.
Andrea Langella works on customer experience and digital engagement strategy in life sciences, focused on how pharma reaches HCPs and patients across channels. Last updated June 2026.