AI Fertility Solutions: What Actually Exists, and What to Require Before You Buy

“AI fertility solutions” is sold as one thing. It is at least five, and they have almost nothing in common.

One of them assesses embryos. One writes notes. One answers the phone. They are bought by different people in the clinic, validated to different standards, and carry entirely different consequences when they get something wrong. A clinic evaluating them as a single purchase is the most common way this goes badly.

This is what the category actually contains, what the evidence supports for each, and what to require before signing.

Five categories, not one technology

Embryo selection and assessment. Software that evaluates embryo images — increasingly time-lapse sequences rather than static morphology — to assist grading and prioritisation. Used by embryologists. Clinically consequential.

Stimulation-protocol optimisation. Tools that suggest dosing or timing based on patient characteristics and cycle response. Used by clinicians. Also clinically consequential, and separately regulated in some jurisdictions.

Clinical documentation. Ambient scribes that capture the consultation and produce a draft note. Used by clinicians, and the change is to time rather than to treatment.

Patient communication. Chatbots handling routine enquiries, and generated video or multilingual explanation. Used by coordinators and front desk.

Analytics and reporting. Computing outcome and laboratory indicators continuously rather than assembling them for a quarterly meeting. Used by leadership and quality.

The differences that matter are not technical. They are who uses it, what happens when it is wrong, and what standard of evidence is appropriate before you rely on it. A scribe that omits a sentence produces an incomplete note a clinician can correct. A tool that influences which embryo is transferred does not offer the same correction opportunity.

Where the evidence is strong, and where it is not

This is the section most vendor material skips, so it is worth being direct — including where it is unflattering to software generally.

Documentation is the best-evidenced category, and the evidence is not uniformly positive. Published multi-platform evaluation shows meaningful error rates in generated notes, with omission rather than fabrication as the dominant failure. That does not argue against the category; it argues for the review step being designed rather than assumed. We covered the findings in AI scribe accuracy.

Analytics is the least contested, because the claim is narrow. Computing an indicator continuously rather than quarterly is an engineering statement, not a clinical one, and it is verifiable in a demo.

Patient communication is straightforward to assess on its own terms — deflection rates, response times, language coverage — because what it claims is operational.

Embryo selection is where claims outrun consensus. The direction of the research is promising and the field is active. But a tool influencing embryo prioritisation needs to state what it was validated against, on which population, and how it performs on cases resembling yours. “Trained on thousands of images” describes the training set, not the performance.

Stimulation optimisation sits in the same category of caution, with the added complication that dosing decisions are the clinician’s responsibility regardless of what a tool suggests.

The pattern is consistent: the categories with the strongest operational evidence make the smallest clinical claims. That is not a criticism of either group — it is the thing to hold in mind when a single vendor presents all five as equally ready.

What you can buy today, and what to keep evaluating

For a clinic deciding where to spend attention this year rather than next:

Deployable now, with a defined review step. Documentation, patient communication, analytics. These are in production across clinics, their benefits are operational and measurable, and the failure modes are recoverable. MedAI Scribe covers documentation, MedXbot handles routine enquiry, MedAI-X covers patient explanation, and AI & Analytics computes the indicators.

Evaluate, but on a clinical standard. Embryo selection and stimulation optimisation. Not because they lack promise, but because the evidence question is harder and the consequence of being wrong is different in kind.

Worth stating plainly, because it is the sort of thing a vendor usually leaves ambiguous: an IVF EMR does not perform embryo selection. Assessment happens on the imaging system. What an EMR does is receive that output and attach it to the right embryo, in the right cycle, in a record that survives. MedART integrates with time-lapse systems including Geri, EmbryoScope+ and Miri TL, syncing annotated footage into the embryo record. That is a real and useful capability, and it is not the same as selecting the embryo. Any vendor blurring that distinction is worth a follow-up question.

What to require before signing

Four questions, and the honest answer to each tells you more than a demo.

1. What was it validated against, and on whose data? A population resembling yours, or a published dataset from a different case mix? For anything touching a clinical decision, this is the whole question.
2. What happens when it is wrong? Not whether it errs — everything does. Whether the workflow catches it, who reviews the output before it reaches a record, and whether that review is recorded.
3. What does my team have to do? Every category carries internal work. A vendor implying otherwise has either done it many times and is simplifying, or has not done it.
4. What does it connect to? A tool producing output that never reaches the clinical record has moved work rather than removed it.

A vendor who answers all four plainly, including the parts that land on your side, is describing something they have deployed. One answering with capability claims is describing a roadmap.

The short version

Treat “AI fertility solutions” as five purchases with five evaluation standards, not one decision.

The operational categories — documentation, communication, analytics — are ready, measurable and worth doing now, provided the review step is designed rather than assumed. The clinical categories deserve more scepticism than enthusiasm, not because they will not arrive, but because the evidence they require has not fully arrived yet.

For where the technology is heading over the next few years rather than what to buy this year, see the state of IVF technology in 2026. For how AI and the clinical record work together once a tool is in place, see AI and EMRs in IVF clinical decision-making.