-
-
About us
Open roles
Product manager
Watch our latest webinar: Who can you trust? A pharma leader’s guide to vetting AI for insights
View Now Product manager
By Dan Gallagher, Senior Consultant, Day One Strategy
AI personas are synthetic models designed to simulate the perspectives of healthcare professionals, patients or other stakeholders using defined evidence, behavioural frameworks and domain knowledge. In pharma they are increasingly used to improve communications, refine messaging and strengthen research before fieldwork begins.
In this article, I cover three things:
Moving from market research into my new role in Digital & AI consultancy has taken me on a steep learning curve of how Generative AI is shaping the pharma marketing and insights space.
The first thing I found a natural affinity for is the idea of synthetic HCP and patient personas.
The value and application makes sense to me. I’ve been on those long, expensive communications testing programmes where we tested too many weak ideas, didn’t get a clear answer, went through multiple rounds and updates to get alignment, and everyone slightly lost the will to live by round three. It’s an inefficient cycle.
So when I first saw AI personas being used to stress-test messaging, refine positioning and sharpen briefs before spending six figures on fieldwork, it made sense.
But, and this is an important but, I am not blindly bought into AI as the answer to everything.
There are pitfalls. And there are plenty of claims about “synthetic audiences” that sound impressive until you ask what is actually sitting underneath them.
Ultimately it’s about understanding where AI can best complement and facilitate us as insights and pharma experts.
So for me, the real opportunity is not simply faster research. It is a better, more insight-driven way of developing communications. One where AI personas are used earlier and more deliberately to improve the quality of what we create, what we test, and what we ask agencies to build.
Why does pharma communications development feel so inefficient?
A lot of pharma communications development still follows a familiar pattern.
A brand team sets a strategic direction. An agency creates ideas and turns them into concepts. Research is commissioned to work out which ones land. The findings come back mixed. The concepts are revised. Sometimes they go back into research again.
Then someone says, “Maybe we need another round.”
At which point everyone smiles politely and quietly updates the timeline.
Often, the issue is that research is being asked to solve problems that could have been tightened earlier:
Does this feel like patient language, specialist language, or something awkwardly stuck in the middle?
These are not always questions that need a full research cycle to explore first.
They are questions where AI personas can help teams think earlier, iterate faster and enter research with stronger options.
That distinction matters. AI personas should not replace the discipline of human research. They should help make sure we are testing better things in the first place.
One of the biggest traps in the current conversation is framing AI personas mainly as synthetic market research respondents.
That can be useful, but it is too narrow.
For me, the more interesting use case is using personas and AI grounded in data and solid rules as a support layer across communications development.
They can help pressure-test whether a message feels credible, where a skeptical HCP might push back, whether patient language feels too soft or too clinical, and whether a concept is actually worth putting into research at all.
So you are not just generating responses and more data. You are refining the work.
In practice, this allows teams to explore alternative message territories earlier, adapt thinking for different segments or customer types, and see which routes feel stronger before committing significant time and budget.
Historically, we often did not have the time or resource to properly develop and test different creative routes for every segment. We had to make sensible compromises. One patient type. One HCP segment. One “best fit” message route. Hopefully good enough.
AI changes this!
It gives us the ability to explore more tailored and segment-specific content earlier in the process. Not final content. Not approved content. But stronger starting points.
Deloitte has also highlighted GenAI’s growing role in marketing content development and personalisation, including storyboarding, copy, visual assets and more tailored customer experiences https://deloitte.wsj.com/cmo/5-trends-to-watch-at-cannes-lions-2024-a8dadf78. For pharma, that does not mean suddenly generating endless content for the sake of it. It means using AI to develop more relevant options earlier, then applying proper human judgement to decide what deserves to move forward.
The biggest change I can see coming is not that pharma teams stop doing research or stop using agencies.
It is that the starting point changes.
With a current client in the immunology space – we developed high fidelity AI personas built based on their customer segmentation. These synthetic representations of customers have become firmly integrated into the client insights, marketing and creative agency workflows.
Instead of moving from a blank brief to creative development to research, we have started to see a more iterative way of working.
First, their AI personas can help shape the strategic problem: the audience unmet need, decision barriers, emotional context and likely objections.
Then they can help generate and refine message territories. Not by magically knowing the “right” answer, but by highlighting where an idea is too generic, too vague, too clever, too disconnected from clinical reality, or simply not clear enough.
Before committing to fieldwork, they also support a pre-research readiness check. This is not MLR. It is not a replacement for proper review. It is simply a way of asking whether we are about to spend money testing something that is clearly not ready or would never get approved.
And the impact is huge, from big campaign development programmes, to small ad hoc questions and messages, the client’s personas have become a critical pressure testing forum. All driven by their data.
There is an efficiency point here too. BCG analysis, reported by Axios, suggested that more than 80% of corporate affairs work can be supported or automated by AI, with teams potentially reclaiming 26%–36% of their time across routine, content-led and data-driven tasks https://www.axios.com/2025/09/18/ai-corporate-affairs-bcg-report. The relevant point for pharma is not “replace everyone and let the robots crack on”. It is that AI can remove some of the repetitive work that slows teams down, allowing humans to spend more time on judgement, creativity and decision-making.
This is where I think the market can get distracted.
A lot of AI persona tools sound impressive in a demo. But as we set out in our practical guide, there are different levels of persona capability, and they are useful for different jobs.
At one end, you have Tier 1: general-purpose LLMs – tools like ChatGPT prompted to act like a stakeholder. Useful for early thinking, but not validated against your market.
Then there is Tier 2: LLM wrappers – general AI grounded in your own data, such as brand materials, advisory board outputs or HCP research. More useful, but still dependent on the quality and freshness of what you feed it.
At the more advanced end, you have Tier 3: custom or small language models – purpose-built around a governed evidence base, with more control, traceability and pharma-specific relevance.
All three can be useful. The point is knowing what level of confidence you need before you use the output to shape a decision.
But in pharma, the real value is not just the model. It is the intelligence wrapped around the model.
A persona is only useful if it knows what to care about, and if output frameworks are relevant and meaningful for pharma insights and marketing teams.
An HCP persona needs to weight clinical evidence differently to a patient persona. A specialist may care more about mechanism, anatomical accuracy and endpoint credibility. A payer lens needs to bring access, value and reimbursement into the conversation. A patient-facing persona needs to understand clarity, tone, risk and emotional sensitivity.
That does not happen by accident.
It comes from the quality of the inputs, the way the model is trained, the templates it works through, and the human expertise guiding how outputs are interpreted.
This is why pharma-grade intelligence matters.
Not just AI capability, but the combination of pharma insight experience, healthcare data understanding, therapy area knowledge, research discipline, strategic judgement and human interpretation.
Without that, AI personas risk becoming polished noise. With it, they become a powerful support layer for better decision-making.
The real promise of AI personas in pharma is not faster outputs.
It is better inputs into the decisions that matter.
Better ideas before research. Better questions during research. Better briefs for agencies. Better clarity before investment.
That is where I think the next phase of value sits.
PS – I am still very much pro human research and pro agency craft. I just think we can stop making all these work quite so hard on ideas that were not ready in the first place.
Day One has pulled together a practical guide on synthetic personas, because the language around them is getting messy. The short version: not all personas are built the same, and not every use case needs the same level of sophistication. Sometimes a simple GPT-style persona is enough to get thinking moving. Sometimes you need something more grounded, controlled and traceable. Download your copy here https://dayonestrategy.com/resources/a-practical-guide-to-ai-personas-in-pharma/
