A self-regulatory advertising body has recommended that Nucleus Genomics modify or discontinue claims that its embryo screening service can accurately rank embryos by disease risk, height and IQ, and Nucleus has said it will appeal

What NAD said it decided, and what it was looking at

This piece is written on the decision summary published by BBB National Programs, not on the full text of the decision, which the same page says is available to subscribers of the online archive. Everything below is what NAD said, in NAD's published account of its own case. The full decision would be the upgrade trigger, and the case is identified only by the advertiser name, Nucleus Genomics, Inc., with no docket number given.

NAD says the matter arose as a monitoring inquiry rather than a competitor challenge, and that it looked at two pieces of advertising: a slideshow on the Nucleus website and an Instagram video. NAD describes the slideshow as opening with the line that you can now see your future child's health, traits and potential before pregnancy even begins, presenting five embryos that all look the same, and culminating in a compare your embryos chart assigning each of the five specific values for cancer, type 2 diabetes and autism spectrum disorder risk, eye and hair colour, height and IQ.

NAD says the Instagram video features the founder of Nucleus asking a woman whether she would want her baby to be taller or smarter, and, referring to six embryos one of which is the smartest, whether she would pick that one.

Five findings follow, on comparative disease risk ranking, on lowest overall cancer risk, on eye and hair colour, on height and on intelligence. In four of the five, NAD says it recommended that Nucleus modify its advertising. On the lowest overall cancer risk claim it says it recommended that Nucleus discontinue the claim or modify it. The composite formula, modify or discontinue, is the one NAD uses in its own summary of the matter as a whole.

The same substantiation gap, five times over

NAD does not say Nucleus had no evidence. On the comparative disease risk claim, NAD says Nucleus supported it with polygenic risk models showing average benefits from selecting lower-risk embryos, and that this evidence "did not establish how accurately IVF+ could rank a full group of five embryos, or validate the percentage-point differences displayed in its advertising".

The height finding repeats the structure. NAD says that while adult height is highly heritable, the proprietary performance measures failed to quantify the accuracy of identifying the tallest embryo in a group of five or six, or to validate the magnitude of the specific inch differences. The intelligence finding does the same: research and polygenic scores for cognitive ability existed, and in NAD's account failed to establish the accuracy of identifying the highest-IQ embryo within a multi-embryo group or to validate the magnitude of the specific IQ-point differences shown.

Two of the five turn on a different defect. On the lowest cancer risk claim, NAD says the statement did not specify a particular cancer, that the evidence concerned specific cancer models, and that it did not explain how separate cancer-specific results are combined into a single overall ranking. On eye and hair colour, NAD says Nucleus provided research on DNA-based predictions but that it failed to show how accurately the models identify the specific labels on the chart, naming green eyes and brown or dark brown hair.

Put the five together and a single proposition falls out, which is ours to state rather than NAD's: the evidence a model developer naturally holds is population-level and aggregate, and the claim an interface naturally makes is individual and specific. Nothing in the summary suggests Nucleus was accused of inventing its research. What NAD says is that the research answered a different question from the one the advertising put to a consumer.

What this is not, and where it goes next

NAD is a self-regulatory body, not a regulator and not a court. Its decisions are not binding law, they carry no civil penalty, and nothing in this decision makes any Nucleus claim unlawful. Any reader who has seen this reported as a ruling against Nucleus should discount that word.

The decision is also contested. NAD's summary records that in its advertiser statement Nucleus stated it "will appeal the NAD's decision" because its "conclusions cannot be reconciled". An appeal from NAD goes to the National Advertising Review Board. We do not state a timetable for that appeal, and the summary gives none.

The one point where a non-binding process acquires weight is compliance. NAD says it will treat the claim Nucleus voluntarily and permanently discontinued during the inquiry, regarding seeing a future child's health, traits and potential prior to pregnancy, for compliance purposes as though NAD had recommended it be discontinued. The summary does not describe what happens if an advertiser declines to follow a recommendation, and we do not fill that gap.

The page carries one more line worth respecting: per NAD and NARB procedures, the release may not be used for promotional purposes. That is a restriction carried on the release itself, and it is a useful signal of how the body treats its own output.

The transferable point for anyone shipping a ranked model output

The product pattern NAD looked at is not unique to reproductive genetics. A model produces scores, an interface sorts a small set of items by those scores, and each item is labelled with a specific number. In our reading the same pattern also appears in credit, hiring, insurance, clinical triage and consumer health products, and we put no figure on how widespread it is.

What the five findings suggest, as our reading of them, is that such an interface makes three claims at once and not one. It claims the underlying model is valid. It claims the model ranks correctly within a small set. And it claims the displayed magnitudes, the percentage points, the inches, the IQ points, mean what they appear to mean. Evidence for the first does not carry the second or the third.

The design consequence is unglamorous. A ranked chart with specific numbers puts a substantiation burden on the numbers. A presentation that expresses the same output as a range, a band or a qualitative signal makes a smaller claim and needs less behind it. That is a choice made in a design review, usually long before legal sees the asset.

None of this is a legal standard. It is what one self-regulatory body said in one contested matter about one advertiser, and a body that reviews advertising is not a body that reviews models. Read it as an early indication of how claim substantiation is being applied to algorithmic outputs, not as a rule.

What we did not verify

What we opened: the BBB National Programs decision summary page, fetched live this session and read in full, including the dateline reading New York, NY, September 11, 2026, all five section findings, the paragraph on the voluntarily discontinued claim, the advertiser statement sentence and the date stamp at the foot of the page.

What we did not open: the full text of the NAD decision, which the page says is available through the BBB National Programs online archive on subscription. We did not read the Nucleus submissions, the polygenic risk models or proprietary performance measures NAD refers to, the advertising itself, any NARB filing, or any NAD and NARB procedures document beyond the line quoted on the page. Because we worked from the body's own summary rather than the operative text, we write throughout that NAD said or recommended, and never that NAD held or that its reasoning was.

What we refuse to claim: we do not say Nucleus advertised unlawfully, deceptively or in breach of any statute, because a self-regulatory recommendation is not a finding of illegality and none was made. We do not say the decision is final, because Nucleus has said it will appeal. We do not say polygenic scores are unreliable or that embryo screening does not work, because NAD made no such finding and we are not equipped to. We do not say what happens if an advertiser declines to comply, because the summary does not say. We give no case number, because the page gives none.

Quotations are reproduced with ASCII apostrophes and quotation marks in place of the typographic characters on the published page, which is a house typesetting convention and not a change to any word.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• The UK ASA on AI advertising disclosure monitoring
• FTC on AI accuracy suppression and section 5 deception
• Korea's KFTC guidance on AI safety claim substantiation
• North America AI regulation tracker
• AI Regulation News hub