There is a real and genuinely unusual body of Harvard epidemiological work behind this post. Since the 1980s, Harvard researchers have collected repeated food-frequency questionnaires and health outcomes from the Nurses' Health Study, Nurses' Health Study II, and Health Professionals Follow-Up Study. In dairy and diabetes analyses from these cohorts, the combined analytic samples total roughly 192,000 people, with follow-up running from 1986 to 2012 and 2013.

In 2018, then doctoral student Andres Ardisson Korat defended a dissertation at the Harvard T.H. Chan School of Public Health analyzing dairy foods and chronic disease. One analysis found that among people with diabetes, eating about half a cup of ice cream per day was associated with a lower risk of cardiovascular problems. Ardisson Korat himself wrote in the thesis that there were "few plausible biological explanations for these results," and the finding was reported as one that researchers could not easily explain away but also did not promote. This is the story The Atlantic's David Merritt Johns reported in 2023, framing it as an uncomfortable and largely unpublicized result in nutrition science rather than a health recommendation.

An inverse association between ice cream intake and type 2 diabetes also appears in earlier Harvard cohort work on dairy and diabetes in men, where ice cream consumption of two or more times per week was reported as associated with reduced diabetes risk. I was able to confirm this only through secondary and advocacy sources, not by retrieving the original table.

On mechanism, the milk fat globule membrane is a real biological structure and a real hypothesis. Coverage of the Atlantic piece describes MFGM as a point sometimes raised by dairy advocates, with some data suggesting dairy products in which the membrane remains intact may be more metabolically neutral than products like butter where churning disrupts it. This is explicitly presented as speculation for future study, not an established explanation. A small 2019 human trial found an MFGM-enriched supplement added to a high-fat meal attenuated postprandial insulin secretion in overweight and obese subjects. That is a short-term, small-sample, surrogate-endpoint finding.

On labeling, the FDA standard of identity requires a product labeled "ice cream" to contain at least 10 percent milkfat, with that fat coming from dairy sources, plus minimum milk solids and weight and air requirements. Products falling below those thresholds, or substituting vegetable oils, must be labeled "frozen dairy dessert." Major brands including Breyers and Dreyer's did reformulate product lines that consequently cannot be called ice cream. I found no data source of any kind quantifying what share of store-brand ice creams fail the standard, and specifically nothing supporting 90 percent.