Who does AI think an IVF patient is?

At the end of 2025, I was preparing a lecture for MSc students on intersectionality and neoliberalism in IVF (basically how different life circumstances shape people’s experience of fertility treatment, and why patients are often held responsible for problems that are caused by the healthcare system).

I wanted the lecture to be visual, so I used generative AI to create some images for my slides. As I generated more images, I found myself thinking, “This can’t be right.”

The patient AI kept creating looked remarkably similar every time. Young. White. Female-presenting. Comfortable. Already portrayed as treatment successful.

What I couldn’t see were the patients I usually treated. Older people. Men. LGBTQ+ people and families. People from different ethnic and cultural backgrounds. People who looked anxious, overwhelmed or uncertain.

So, when I was talking to Katie Rollings from the charity Fertility Action about what I’d been seeing, we both agreed it was worth looking at more closely. What started as a conversation soon became a research project.

As Katie and I talked about what I had been seeing, we both started asking the same question. Was this simply one AI platform producing odd results, or was something much bigger going on?

The only way to answer that was to look at it systematically. We decided to use the same prompt — “Generate an image of an IVF patient in a fertility clinic” — across several of the world’s most widely used AI image-generation platforms. We deliberately kept the prompt simple because we wanted to understand what these systems produced by default, without steering them in any direction.

We also wanted the study to reflect how people use generative AI. Rather than changing lots of settings or creating complex prompts, we used the platforms in exactly the way most people would: on a standard smartphone using their default settings. We generated images across multiple platforms, repeated the process several times and used accounts registered in different countries, including both Western countries and the Middle East.

Once we had collected the images, we analysed them using qualitative visual content analysis. We weren’t interested in which platform produced the nicest-looking image or the most realistic clinic. We wanted to understand the story the images were telling about who an IVF patient is. We looked at characteristics such as age, gender presentation, ethnicity, body type and appearance, but also at the emotional tone of the images, the stage of treatment they appeared to represent, the clinical environment and — perhaps most importantly — who was missing altogether.

By comparing the images across different platforms, countries and repeated generations, we wanted to see whether the differences outweighed the similarities, or whether the same assumptions kept appearing regardless of which AI system we used.

The results

I expected to see some similarities, but I also expected each platform to have its own “personality”. I thought there would be cultural differences too. Surely an AI account based in the Middle East would generate different images from one based in the UK or Europe.

It didn’t.

What surprised me most wasn’t the differences. It was the consistency.

Across every platform we tested, the images were remarkably similar. The “IVF patient” was almost always portrayed as a young, female-presenting, conventionally attractive, able-bodied woman in a calm, modern fertility clinic. She appeared financially comfortable, emotionally reassured and already well on the way to achieving a successful outcome.

The details varied slightly from platform to platform. Some focused more on pregnancy, others on the clinical consultation, while others created almost lifestyle-style images of the “journey to motherhood”. But underneath those stylistic differences, the same assumptions kept appearing.

More striking was what we didn’t see.

None of the platforms generated male patients, LGBTQ+ people or older people. There was very little ethnic diversity, despite generating images using accounts registered in different countries and cultural contexts. We found no representation of people living with disabilities.

Equally absent were many of the experiences that define fertility treatment for so many people: repeated treatment cycles, treatment burden, uncertainty and emotional distress.

Even the fertility clinic itself looked remarkably similar across platforms. Bright, modern, private and reassuring. The technology was there, but mostly as a visual symbol rather than as something patients actually interacted with.

One finding stayed with me more than any other. Even when we generated images using AI accounts based in the Middle East, the outputs were identical to those created in the UK and Europe. Rather than reflecting the cultural diversity of the populations they were supposedly serving, the systems continued to produce the same Western-looking, idealised patient. I genuinely hadn’t expected that, and it made me realise that these default representations weren’t simply platform specific. They appeared to be remarkably consistent across AI systems and across different parts of the world.

By the end of the analysis, it became clear that this wasn’t about one image generator getting it wrong. Different platforms produced different styles of image, but they all told essentially the same story about who an IVF patient is — and perhaps more importantly, who they are not.

What does this mean for fertility care?

At first glance, it would be easy to dismiss this as a study about AI-generated images. I don’t think it is.

Generative AI is already being used throughout fertility care. Clinics are using it to create websites, patient information, educational resources and digital communications. As these technologies become more integrated into healthcare, the images and messages they generate will increasingly shape how patients see fertility treatment — and how fertility treatment sees them.

If AI repeatedly presents one narrow version of an IVF patient, there is a real risk that everyone else becomes invisible. Not because they don’t exist, but because they aren’t represented. Older people, men, LGBTQ+ people, people from different ethnic backgrounds, people living with disabilities, and those experiencing repeated unsuccessful fertility treatment may never see themselves reflected in the information designed to support them.

Representation is not just about creating inclusive images. It is about recognising who belongs, whose experiences matter and whose needs are considered when we design the future of fertility care.

Technology has enormous potential to improve the patient experience, but only if it reflects the reality of the people it is intended to serve.

Because if patients cannot see themselves, perhaps the more important question is this:

How do we treat patients we can’t see?

The full research abstract, “Bias by default: a qualitative visual cross-platform analysis of how generative AI models construct the IVF patient through problematic default representations worldwide,” is published in Human Reproduction and is available here:

Read the full abstract in Human Reproduction.