THE PGT-A CONTROVERSY: What Five Years Later A.I. May or May Not Change
Today’s posting offers an article by Lyka Mochizuki, MSc, who, several years ago while a research assistant and research fellow at the CHR, wrote a remarkable article on PGT-A in the JARG and now – here – is revisiting the same theme. But without disclosing too much, it wasn’t then just another opinion article on PGT-A and it is not now either, and you will have to read the article to know why. We think you will find the article to be quite interesting.
And if you wondered why we so far didn’t post this week, Monday was, of course, a holiday and Tuesday we were extremely busy completing the April/May issue of the CHRVOICE, which is scheduled to drop tomorrow.
The CHR’s Editorial Staff
THE PGT-A CONTROVERSY What Five Years Later A.I. May or May Not Change
By: Lyka Mochizuki, MSc, past Research Fellow and currently Project Manager at the CHR. She can be reached through the editorial office of the CHRVOICE and The Reproductive Times.
This article follows a highly unusual article on the then still very acutely controversial subject of preimplantation genetic testing for aneuploidy (PGT-A) in association with in vitro fertilization (IVF) by the same author in the Journal for Assisted Reproduction and Genetics. The article was so unusual because in addressing the then still much more heated differences of opinion, she for the first time applied the concept of conflict resolution to the dispute (the author holds a MSc in Conflict Resolution from Columbia University in NYC). Now - in a different position - once again a member of the CHR team, she agreed to revisit the subject, though in consideration of the current Zeitgeist in which A.I. is changing the world and, with it, of course, medicine. We are certain you will enjoy this article and, as always, we are welcoming responses.
And as a side note, the April/May CHRVOICE should be dropping tomorrow! Here, too, we are awaiting your responses.
In 2020, I co-authored a paper in the Journal of Assisted Reproduction and Genetics that approached the controversy surrounding preimplantation genetic testing for aneuploidy (PGT-A) through an unusual lens: a formal conflict resolution analysis.¹ Rather than entering the clinical debate as another voice for or against PGT-A, the paper examined why the debate itself had become so durable — and what it would take to move past it.
The argument was that the persistent disagreement over PGT-A could not be resolved by more data alone. It was a structured conflict between physicians and professional bodies, rooted in competing interpretations of overlapping evidence, competing incentives, and competing definitions of what counts as a successful in vitro fertilization (IVF) cycle.
Five years later, the controversy has evolved, but not resolved, and a new element has entered the conversation, - artificial intelligence (A.I.) applied directly to PGT-A. This article, therefore, asks two questions: Can A.I. resolve the conflict between physicians, or relocate it? And what would real resolution actually require?
Where the Controversy Stands in 2026
The clinical evidence has not been kind to the broadest claims made for PGT-A: A 2021 randomized controlled trial in the New England Journal of Medicine found that, among good-prognosis patients, cumulative live birth rates were lower in the PGT-A group than in the conventional IVF group.² A 2025 review described PGT-A’s clinical value as an open controversy after nearly thirty years of practice.³ A recent analysis of the UK Human Fertilisation and Embryology Authority registry found that, in routine practice, PGT-A was associated with reduced live birth rates compared to standard IVF.⁴ The American Society for Reproductive Medicine’s 2024 committee opinion concluded that PGT-A should not be recommended for routine use.⁵
At the same time, subgroup evidence has somewhat shrouded the picture when some authors suggested that PGT-A improves live birth rates in women over age 35,⁶ though ignoring the growing importance of patient selection bias with advancing female age. Successful blastocyst-stage culture – an obvious prerequisite for PGT-A – of course increasingly biases patient selection toward good-prognosis patients, as poorer-prognosis patients simply will no longer produce blastocysts. Obstetrical safety data after PGT-A use are, however, reassuring.⁷
That ambiguity has become the controversy. Physicians who emphasize per-transfer success and argue that avoidance of miscarriages is a worthwhile goal to pursue read this evidence one way. Physicians who instead emphasize cumulative outcomes and embryo conservation read it in a very different way. Both sides claim not to misread the data and my comments here are not meant to be the judge between these two opinions (even though – as a scientist – I, of course, have an opinion). What I am instead trying to achieve, is to stop both sides from yelling at each other.
What A.I. is Doing Now
A.I. has attempted to move from embryo morphology into the genetic analysis itself. A 2023 study of nearly 25,000 embryos reported that A.I.-augmented PGT-A allegedly produced higher euploidy rates, lower mosaicism rates, and was associated with higher live birth rates in subsequent transfers.⁸ [Please note industry conflict regarding this statement further explained at the end of this manuscript.] Other A.I. models are alleged to detect complex abnormalities, including triploidy, that standard PGT-A can miss.⁹ Interestingly, the MIT Technology Review named A.I.-based embryo scoring one of its breakthrough technologies of 2026.¹⁰
While the technical promise appears to be real, - reliability was, however, never the whole controversy. It was always only one layer of it.
The Layers A.I. Touches, and Doesn’t Touch
It is reasonable to assume that the PGT-A controversy has three layers and that A.I. affects them unevenly. The first is technical and is defined by how accurately can the test classify an embryo’s chromosomal status? This is likely the layer A.I. can most directly improve.
The second layer is clinical: Even when the classification is accurate, does acting on this information really improve outcomes across an entire IVF cycle? This is where most of the disagreement between physicians lives. While A.I. can improve the label, it does not necessarily resolve the question of what to do with the label. This, of course, especially applies to mosaic or intermediate results, where clinical judgment remains shaped by individual values and risk tolerance.
The third layer is structural and addresses the question, - what is PGT-A for, and who benefits from its widespread use? Somewhat surprisingly, adoption has continued to grow, - even as evidence for its routine use has weakened. The financial structure of IVF adds additional complexity: So-called “add-on” tests are often billed outside of insurance coverage. PGT-A is a good example: Considering PGT-A is still experimental, insurance companies even in covered IVF cycles usually do not include PGT-A coverage. Patients, therefore, pay for this service out of pocket and undiscounted, greatly enhancing overall cycle profitability for IVF clinics, otherwise forced to greatly discount cycle fees to insurance companies.
Moreover, while advertising IVF outcomes with reference embryo transfer rather than cycle start, success rates offered to the public are often significantly exaggerated and unrelated to increasingly accepted evidence. A.I. does not touch this layer but – if anything – A.I.-augmented PGT-A may, in itself, become a new add-on with its own opportunity for marketing.
Our 2020 paper argued that more data alone likely could not resolve this conflict and A.I. in effect means more data. It is improving the technical layer while leaving the clinical and structural layers untouched. The conflict has not been resolved, but its terrain has shifted.
Toward Common Ground
If A.I. cannot resolve the disagreement between physicians, then what can? We suggest that three steps, drawn from conflict resolution practice rather than clinical debate may be able to achieve this goal.
Agree on the outcome that matters most. Much of the disagreement disappears when physicians specify, in advance, whether they are optimizing for per-transfer success, cumulative live birth per cycle started, time to first pregnancy, or miscarriage reduction. These are different goals, and a test that helps one can hurt another. Naming the goal before recommending the test is the single most useful step.
Separate the test from the decision. A.I. improves PGT-A’s classification accuracy, but classification is not the same as a transfer decision. A mosaic or intermediate result is a piece of information, not an instruction. Building decision frameworks that treat ambiguous results as starting points for patient conversation, rather than disposal criteria, would address one of the most persistent sources of physician disagreement — and one of the most painful sources of patient harm.
Acknowledge the structural layer openly. The financial and competitive structure of IVF shapes which tests get offered, marketed, and adopted. Pretending those incentives do not exist makes the clinical debate harder than it needs to be. A field that openly discusses how its payment structure influences its practice patterns is better positioned to evaluate new technologies on their merits, — including A.I.
None of these steps requires a new study. They require a different kind of conversation, between physicians who currently talk past each other.
Why This Matters Beyond PGT-A
This behavior pattern is likely not unique to reproductive medicine. Across health care, A.I. is being introduced into long-standing controversies, such as mammography screening, prostate cancer overdiagnosis, and even psychiatric diagnostic categories. In each case, A.I. offers improved classifications, better consistency, and faster analysis. And in each case, the underlying disagreement persists, because tests do not exist in a vacuum. They exist inside workflows, payment structures, and definitions of success that carry their own incentives.
A more accurate test inside a contested system produces more accurate contested results. Resolving a conflict between physicians requires more than better information. It requires a willingness to revisit the goals, the incentives, and the values that shaped the disagreement in the first place.
This work, for now, at least remains in human hands.
A.I. Disclosure Notice:
This article was written with help of several A.I. platforms during the research phase. The final product was fully vetted and edited.
Industry Conflict:
This study⁸ was conducted at NYU Langone Prelude Fertility Center using CooperSurgical’s proprietary PGTai℠ and PGTai 2.0 platforms. CooperSurgical has a direct commercial interest in PGT-A. This article acknowledges that the strongest evidence for A.I.-augmented PGT-A benefits has come from studies on commercial A.I. platforms by groups with industry relationships, an observation that strengthens the article’s central argument that A.I. does not resolve the underlying conflict, but only relocates it.
REFERENCES
Mochizuki L, Gleicher N.. J Assist Reprod Genet. 2020;37(3):677-687.
Yan J, et al. N Engl J of Medi. 2021; 385(22):2047-2058
Coussa et al., J Assist Reprod Genet. 2025;42:179-184
Roberts, et al., Reprod Biomed Online. May 2, 2026:105747. S1472648326002889
ASRM; A Committee Opinion 2024. Fertil Steril. 2024;122(3):421-434
Simopoulou et al. J Assist Reprod Genet. 2021 38(8):1939-1957
Hyttel, et al. Hum Reprod. 2026;deag049; on line ahead of print.
Buldo-Licciardi et al., J Assist Reprod Genet. 2023;40(2):289-299
Xiet et al., Hum Reprod Open. 2025;,hoaf054
MIT Technology Review. May 7, 2026. https://www.technologyreview.com/2026/05/07/1136946/whats-next-for-ivf-ai-robot-pgt-gene-editing/