Are you adapting your scientific communications to the AI era? A guide for pharma

By Magdalene Karipidou, Digital Manager and Maddy Parker, Associate Principal Medical Writer

What does AI have to do with publications and medical communications? 

AI-generated answers are no longer a fringe behaviour: 78% of clinicians turn to ChatGPT for professional use, and many are experimenting with platforms like Gemini, OpenEvidence and Claude. 

Patients, payers, journalists and analysts rely on AI answers too.

Unlike a traditional search, where users click through to a source and make their own judgement, AI tools synthesise and summarise information from multiple sources, meaning that the answer itself becomes the endpoint for many users. This may not need to be through a dedicated application, but via an AI-generated summary built into the search engine output, like Google’s AI Overviews.

It’s not all about AI visibility in the world of healthcare

Generative engine optimisation (GEO), or Answer Engine Optimisation (AEO), has arrived in pharma, and most of the early conversation has focused on visibility.

‘How do we get our content cited?’, ‘How do we show up in AI-generated answers?’, ‘How do we increase our presence across tools like Perplexity and Google’s AI Overviews?’. These are the questions driving most of the discussion. They make sense, because GEO and AEO are terms that come from the broader marketing world, where the goal is straightforward: use tactics that increase the likelihood that a brand, organisation or piece of content is surfaced and cited by generative AI tools.

But pharma is different. The stakes are higher.

In healthcare communications, the conversation shouldn’t start with visibility, but with accuracy. Specifically, we should start with improving how scientific evidence is identified, interpreted and referenced by these models.

If AI tools are surfacing your evidence inaccurately, then optimising for visibility just means that more people see the wrong information. In a sector like pharma, a misrepresented safety profile or an outdated efficacy claim can impact patient care. Misinformation could carry regulatory, clinical and reputational consequences. Thus, in healthcare, accuracy must come first.

The accuracy problem is bigger than many people realise

Research published in Nature Communications in 2025 found that 50–90% of large language model (or LLM) responses were not fully supported – and were sometimes contradicted – by the sources cited. Even web-enabled models left substantial gaps between answer and evidence.

This is not a technology problem that will resolve itself, but a structural challenge that communications and publications teams need to actively manage.

The issue is compounded by how these tools work. They don’t distinguish between your medical information page and a press release from a data readout three years ago. They pull from whatever is accessible – publications, trial registries, prescribing information, disease education content, congress abstracts, owned websites, social media and third-party summaries – and generate an answer from the aggregate. Simply put, AI is pulling information from the entire digital content ecosystem – and not always from your own content. If the most credible, current and QC-vetted material is less visible or harder to interpret than third-party summaries, older press releases or low-quality commentary, those sources can end up shaping the answer instead.

Most digital ecosystems weren’t built for AI

Publications teams, medical affairs, digital, and corporate communications each manage their own content with their own processes and timelines. That’s a reasonable way to organise the work internally, but it creates inconsistency at the ecosystem level and AI tools will find those inconsistencies.

For example:

• If your trial registry uses different terminology to your publication, ambiguity can compound

• If your most retrievable content is an old press release rather than your most recent medical information, that shapes what gets surfaced

• If a plain-language summary isn’t developed, a third-party simplification that you have no control over may fill the gap instead

• If your publications are written in dense, complex sentences, AI tools are more likely to misrepresent or lose the context of your data

How should your organisation approach this challenge

The practical starting point is an audit, followed by the content plan. Before asking how to increase AI visibility, you need to understand how your evidence is currently being represented.

Start by running realistic queries across relevant tools from the perspective of your key audiences to assess whether the outputs are accurate, current and appropriately contextualised.

What does a meaningful audit look at?

1. Consistency across the digital ecosystem – do publications, labels, trial registries, medical information pages, disease education and owned content align in terminology and claims? Are there contradictions or gaps that create room for misrepresentation?
2. Source quality and retrievability – which assets are being surfaced? Is key evidence accessible, structured and findable? Are lower-quality sources carrying more weight than they should?
3. Fidelity of AI outputs – when AI tools generate answers about your evidence, do those answers explain the data in the correct context? Or are they generating the wrong conclusions?
4. Recency – is current evidence surfacing, or are AI tools drawing on outdated publications and reviews?

The output of that audit should be a prioritised remediation plan. Some actions will sit with:

• Publications – open access, structured abstracts, plain-language summaries

• Med Info – ensuring pages are current, navigable and clearly structured

• Digital – metadata, site architecture, content governance

• Corporate communications – reviewing what older content is highly retrievable and whether it accurately reflects the current evidence base

GEO requires an integrated team

GEO for pharma isn’t solely a digital tactics problem, and it isn’t a publications problem. It sits across all pharma communications, including medical writing, medical affairs, legal, regulatory, public relations and clinical development. 

Optimising the evidence ecosystem requires people who understand the science, people who understand how AI tools retrieve and synthesise content, and people who can write clearly for both human and machine-mediated discovery. That combination is rare in any single function. 

At Madano, we have integrated our insights, digital and medical writing teams. We combine our expertise to tackle this complex challenge. 

If you’re interested to understand how your evidence is being represented across AI-mediated touchpoints – and what to do about it – get in touch, [email protected].