We talk a lot about what AI could do for fertility care. Assess embryos. Predict outcomes. Help clinicians make better decisions. And I understand the excitement, because the potential is enormous.
But before we get too carried away, I think we need to ask a slightly uncomfortable question: where will AI get the truth from?
Consider one fertility journey. For the patient, it is one connected experience. Their history informs the treatment plan. Medication leads to scans and blood tests. Those results shape decisions about dosage and timing. Retrieval leads to lab work, embryo development and perhaps a transfer or cryostorage. Conversations with the patient run through all of it.
The digital version of that same journey looks very different. The clinical record might be in an EMR, embryo information in a lab system, images somewhere else, conversations in a patient portal, results in PDFs and important details in spreadsheets. Each system records the part it was built to handle. Very few can show the whole journey, including how one decision led to the next.
We see the consequences when we migrate a clinic’s data. Sometimes even “How many IVF cycles did you perform last year?” takes real work to answer. The clinic has data, of course. But different systems use different definitions, and their records do not always agree. AI will not resolve that ambiguity simply because we give it access to everything. It will inherit it.
The bigger problem is what those records leave out.
Imagine a medication dose changes from 225 IU to 300 IU. The change is recorded, along with the patient’s results and, eventually, the outcome. But why did the clinician change it? Was it the hormone result, follicle growth, the response in a previous cycle, the clinic’s protocol or something discussed with the patient? What did the clinician know at that moment, and which of those things mattered to the decision?
A dataset containing thousands of dose changes might look impressive. But if it cannot distinguish between those reasons, what exactly are we asking AI to learn from it?
That is why I think we need to change the question. We tend to ask what data we already have and what AI might find in it. We should also ask what we want to understand in the future, then work backwards: what would we need to know about each treatment, decision and outcome to answer it?
For the dose change, that means connecting what happened to what was known, what was considered, why a decision was made and what followed. The same applies in the lab, in conversations with patients and in the operational decisions that shape their care. Those connections are the dataset we need to start building.
I know that can sound like more documentation for already busy teams. It cannot be. This is where AI becomes interesting for a second reason: it can help create the data it needs.
Much of the first generation of fertility AI looks at an input, such as an embryo image or an ultrasound, and produces an assessment or prediction. That work matters, but the next generation can participate across the journey. Alongside a clinician, AI could bring together the relevant history, results and protocol when a dose is being reviewed. It could help document the reason for the eventual decision, with the clinician checking and owning that record. Later, the team could see both what happened and why.
The same principle could help an embryologist connect a lab decision to its context, a coordinator keep track of what was agreed with a patient, or a clinic understand where its usual process is being changed and why. AI becomes useful in the work itself, while helping us build a better record of that work.
This is why the shift from prediction to participation could be so significant for fertility. A fertility journey is not one image or one outcome. It is hundreds of decisions made by different people over time, each influenced by what came before. If AI is going to help across that journey, it needs to understand those connections. And if it participates thoughtfully, it can help us capture them without turning care into a data entry exercise.
We cannot build that future by only cleaning up the records we already have. We need to decide what we want fertility care to be able to learn, and start creating the dataset that makes it possible.
AI cannot learn from what we do not capture. But it might finally help us capture what matters.
Next:
Article
