Seed oils — soybean, canola, sunflower and corn — are the dominant fats in the modern industrial food supply. Soybean oil alone accounted for 80 to 90 percent of US edible oil consumption in 1998, a position attributed to its availability and functional properties rather than to any health rationale. ¹ These same oils appear overwhelmingly inside ultra-processed foods (UPFs), which now supply more than half of total dietary energy in the United States, United Kingdom, Canada and Australia, a figure corroborated independently by the FAO. ¹

That coincidence generates the central question: when chronic disease tracks with seed-oil intake, is the oil the culprit, or the ultra-processed package it rides in? The debate matters because the two explanations point to opposite interventions. If the oils are intrinsically harmful, the remedy is to return to animal fats or tropical oils; if the processing is the problem, seed oils are bystanders and the target is the formulation of the food itself. This report tests each explanation against the graded evidence, distinguishing throughout between what controlled trials show and what population correlations can and cannot establish.

The highest-control evidence — trials that dictate what people eat — points consistently toward benefit. A network meta-analysis of 54 randomized trials with 2,065 participants found that seed oils (sunflower, rapeseed, corn, soybean) reduced LDL-C by 0.42 to 0.23 mmol/L compared with butter in isocaloric exchange. ¹ A single-oil trial reported within a 2025 systematic review is consistent with that range: Jenkins et al. found 31 g of canola oil per day for three months in 141 adults with type 2 diabetes lowered LDL-C by 0.20 mmol/L, alongside a 5.15 mmol/mol drop in HbA1c and a 0.30 mmol/L fall in total cholesterol. ¹ The units are directly comparable, and the canola point estimate falls just inside the pooled seed-oil range — a convergence between a single trial and a 54-trial synthesis. That same systematic review, drawing on 11 studies, concluded seed oils can positively influence lipid profiles and glycemic control. ¹

The official record concurs. The USDA's Nutrition Evidence Systematic Review graded the association between n-6 PUFA and improved blood lipids 'Strong' and concluded PUFA intake significantly lowers cardiovascular and type 2 diabetes risk. ¹ The 2015 Dietary Guidelines Advisory Committee reported that replacing 1% of energy from saturated fat with PUFA reduces coronary heart disease incidence by 2 to 3 percent, with the Farvid 2014 meta-analysis (13 studies, 310,602 individuals) finding a 5% linoleic-acid increment replacing saturated fat lowered CHD events 9% and CHD deaths 13%, and Mozaffarian 2010 (8 RCTs, 13,614 participants) finding a 19% reduction in myocardial infarction or CHD death. ¹

The oxidative-stress harm that seed-oil critics propose does not appear under controlled testing. EUFIC reports that randomized trials consistently show no significant differences in oxidative-stress markers from seed oils, and that a rapeseed-oil diet did not increase oxidative damage relative to a saturated-fat diet; a cited six-week RCT giving 30 healthy young women 15 g/day or 7.5 g/day linoleic acid versus a palmitic-acid control found no significant change in oxidative-stress markers. ¹ These two claims come from a single industry-funded body — EUFIC is documented as an industry-funded organization — and the second is graded REPORTED, resting on that one interested source. ¹ The reader should weight them accordingly, but their direction aligns with the independent lipid evidence.

Open: What baseline HbA1c underlay the Jenkins 5.15 mmol/mol reduction, which the analyst could bound only to roughly a 7.4–7.9% relative fall depending on assumed baseline?; Do the EUFIC-reported oxidative-stress RCTs hold up against non-industry replication with the underlying primary publications?

The single most discriminating piece of evidence comes from an inpatient crossover trial that isolates processing from composition. Cell Metabolism reports that 20 weight-stable adults consumed 508 ± 106 kcal/day more on an ultra-processed diet than on an unprocessed one, even though both were matched for calories, energy density, macronutrients, sugar, sodium and fibre; participants gained 0.9 ± 0.3 kg on the ultra-processed diet and lost 0.9 ± 0.3 kg on the unprocessed one. ¹ The symmetry of that weight change — identical magnitude, opposite sign, same standard error — is exactly what a crossover design predicts. Two further origins describe the same study: the FAO reports the 508 kcal figure with an 81.3% ultra-processed energy exposure, and a Frontiers in Nutrition review reports the effect as roughly 500 kcal. ² These three independent citations converge on an identical point estimate, with the Frontiers figure simply rounded.

Mechanistically, several pathways operate independently of fat type. A randomized double-blind controlled-feeding study found 15 g/day of the emulsifier carboxymethylcellulose for 14 days increased postprandial abdominal discomfort and reduced gut microbiota richness. ¹ A review reports that emulsifiers such as CMC and polysorbate 80 erode the intestinal mucus barrier and increase gut permeability independent of macronutrient content. ¹ Separately, the NutriNet-Santé cohort of 105,588 participants linked the artificial sweeteners aspartame, acesulfame-K and sucralose to higher type 2 diabetes risk, with hazard ratios of 1.69, 1.63 and 1.70 respectively — additives, not fats. ¹

One caveat runs against over-reading the mechanism. The CMC trial found that while microbiota richness fell, fecal lipopolysaccharide and flagellin were not affected, contrasting with the stronger effects seen in animal and in vitro models. ¹ That human/animal gap, and the fact that the human CMC data rest on a single trial, mean the emulsifier pathway is demonstrated more clearly for microbiota composition than for the downstream inflammatory markers critics invoke. ¹

Open: Does emulsifier-driven microbiota disruption in humans translate to the inflammatory endpoints (LPS, flagellin) that animal models predict but the Chassaing trial did not detect?; Would an ultra-processed diet built with saturated fats rather than seed oils produce the same 508 kcal excess, a comparison no cited trial has run?

The most direct test of the rival explanations would be populations with high seed-oil intake but low ultra-processed intake. The claims here are the thinnest in the report, and the direction they point is precisely why that matters. Deutsche Welle, reporting an Oswaldo Cruz Foundation study of eight countries, states that Latin American countries had the lowest ultra-processed dietary share and the lowest ultra-processed-attributable premature deaths, and that Brazil's traditional diet of rice, beans, salad, fruits and protein was described as cardioprotective — a whole-food pattern that nonetheless uses vegetable oil. ¹ That claim is CONFIRMED.

Older population signals are graded lower. The Montreal Heart Institute Prevention Observatory reports a 1970s finding that Japanese emigrants to California had twice the coronary heart disease incidence of those remaining in Japan, pointing to a non-genetic dietary or lifestyle cause, and a Japanese study it cites reported a 45% lower myocardial infarction risk in women with the highest soy intake. ¹ Both are REPORTED from a single advocacy source. ¹ Similarly, MedStar Health reports the PREDIMED trial of about 7,500 patients found Mediterranean-diet participants 30% less likely to develop heart disease — again a single hospital-blog source, graded REPORTED. ¹

One higher-grade olive-oil finding is itself CONTESTED. The European Journal of Clinical Nutrition reports virgin olive oil associated with a 57% reduction in cardiovascular mortality (HR 0.43; CI 0.20-0.91), while refined olive oil showed no association (HR 0.88; CI 0.49-1.60). ¹ The hazard ratios match the source, but the claim describes the comparison as 'extreme tertiles' whereas the paper describes it as negligible consumption versus roughly 20 g/day — a discrepancy left unresolved. ¹ The net effect is that the sub-question most capable of separating oil from processing is answered mainly by single-source, lower-tier claims, and no cited study stratifies hard health outcomes by the source of the oil.

Open: Do the Japanese emigrant and PREDIMED findings survive verification against their primary publications rather than the single advocacy and hospital sources reporting them?; On what exact contrast (tertiles versus a ~20 g/day threshold) does the 57% virgin-olive-oil mortality reduction rest?

The decisive statistical question is whether ultra-processed food's disease associations can be explained away by the nutrients it carries. Two large cohorts say no. The BMJ's NutriNet-Santé study found that a 10% absolute increment in ultra-processed food was associated with 12% higher overall cardiovascular risk (HR 1.12, 95% CI 1.05-1.20), with the association remaining significant after adjustment for saturated fatty acids, sodium, sugar and dietary fibre — arithmetic the analyst confirmed (1.12 minus 1.00 equals a 12% increase). ¹ The Framingham Offspring analysis of 3,003 adults over about 18 years found each additional daily serving of ultra-processed food associated with 7% higher hard cardiovascular risk and 9% higher hard CHD, robust after adjusting for total energy, waist circumference, BMI and diet quality. ¹ These use different metrics — energy increment versus servings — so they corroborate the pattern rather than the exact number.

The correlation structure inside the food supply cuts sharply against the seed-oil hypothesis. A meta-analysis found each 1% increase in energy from ultra-processed food was linearly correlated with about a 0.2 percentage-point rise in added sugars but a ~0.06 percentage-point decrease in plant oils. ¹ If seed oils drove UPF harm, they should rise with ultra-processing; instead they fall. Reinforcing this, in the US only 7.4% of total linoleic-acid intake comes from salad dressing, while about 38.5% comes from low-nutrient, calorically dense processed foods such as chips, pizza, desserts and fried foods — placing linoleic acid inside the processed matrix, but as a marker of that matrix rather than an established cause. ¹

The long-run cohort evidence on linoleic acid itself leans protective: a 2020 meta-analysis of 44 prospective cohorts (811,069 people) found higher linoleic-acid intake associated with lower cardiovascular mortality (RR 0.87), all-cause mortality (RR 0.87) and cancer mortality (RR 0.78). ¹ One internal wrinkle warrants flagging: the Framingham endpoint ordering — hard CHD 9%, hard CVD 7%, overall CVD 5% — is counterintuitive, since hard endpoints are usually a subset of the broader category, and the analyst could not verify the hierarchy without the original definitions. ¹

Open: Why does the Framingham per-serving hierarchy place hard CHD (9%) above overall CVD (5%), and does the original endpoint definition resolve the apparent inversion?; Because C-020's per-10%-energy metric has no independent replicate, how robust is the exact 12% figure beyond the single NutriNet-Santé cohort?

The inconvenient sub-question — are seed oils independently selected for processed foods, or simply the cheapest fat available — is answered mostly by industry and technical literature. Oil selection depends on regional availability, culinary traditions and economic factors, with soybean oil the most common in the Americas and widely used in Asia. ¹ The ICCT reports US mean prices from 1992-2016 of $586/Mt for palm oil, $658/Mt for soybean oil and $774/Mt for canola oil, with palm oil tending to be cheaper than soy. ¹ Soybean oil's 1998 dominance — 80 to 90 percent of US edible oil — is attributed explicitly to availability and functional properties. ¹

That same functional profile complicates any clean separation of 'oil' from 'processing.' Gunstone's technical review notes soybean oil's high linoleic-acid content makes it oxidatively unstable, requiring hydrogenation or modification for many food uses. ¹ In bakery reformulation, the aim is to replace saturated fats with plant-based fats and seed oils, but reducing saturated fat can harm sensory qualities — a constraint that pushes manufacturers toward modification and additives regardless of the base oil. ¹ The palatability layer sits atop this: FoodTimes reports that Fazzino et al. found about 62% of foods in a US database met criteria for hyper-palatability through fat/sodium, fat/sugar or carbohydrate/sodium combinations, potentially increasing consumption by up to 30%, though this is a single REPORTED source. ¹

The economic account explains why seed oils are ubiquitous but does not, on its own, establish that their presence is harmless. Two time caveats apply: the 1998 dominance figure and the 1992-2016 price series predate the 2019 Hall trial by up to two decades, so current market shares cannot be assumed identical. ² The evidence here establishes motive for seed oils' prevalence, not exoneration of their effects.

Open: Has soybean oil's 80–90% share of US edible oil persisted from 1998 to the present, or shifted with newer oils and reformulation?; Does the ~62% hyper-palatability figure survive verification against the primary Fazzino 2019 study rather than the single FoodTimes report?

H1 — seed oils harm health independent of the ultra-processed matrix — has weak standing. If it were true, controlled trials feeding seed oils in minimally processed contexts would show harm to lipids, inflammation or outcomes. Instead, the network meta-analysis of 54 trials shows LDL-C reduction versus butter, the USDA grades the lipid benefit 'Strong', and controlled oxidative-stress testing shows no increase. ³ The hypothesis rests almost entirely on the Sydney Diet Heart Study, which found replacing saturated fat with linoleic acid increased all-cause mortality (HR 1.62, 95% CI 1.00-2.64), with its confidence interval touching 1.00, and on null findings — Chowdhury 2014's RR 1.01 for n-6 PUFA and a 9-RCT omega-6 supplementation meta-analysis at RR 0.94 (95% CI 0.77-1.15). ² Null effects are not positive harm; a single contradicted RCT against pooled data on 811,069 people showing protection (RR 0.87) leaves H1 poorly supported. ¹

H2 — ultra-processing harms health independent of fat type — is the best supported. Its prediction is that harm appears when processing varies but composition is held constant, and that population associations survive nutrient adjustment. Both hold: the Hall crossover produced 508 kcal/day excess intake and weight gain on macronutrient-matched diets, NutriNet-Santé's CVD association survived saturated-fat, sodium, sugar and fibre adjustment, and Framingham's survived energy, BMI and diet-quality adjustment. ⁴ No cited claim contradicts it.

H3 — seed-oil harms are mediated by their presence in UPFs rather than intrinsic toxicity — is plausible and partly evidenced. Its signature prediction is that seed oils rise with processing; the data show the opposite, a 0.05 percentage-point decrease per 1% UPF energy, which weakens the 'seed oils are the vehicle' framing while leaving intact the broader point that only 7.4% of US linoleic acid comes from a whole-food source like salad dressing. ² The population evidence that would clinch H3 — favourable outcomes where seed oils are eaten in whole-food contexts — comes largely from REPORTED single-source claims, so H3 remains supported in direction but under-evidenced. ²

H4 — additive independent harm from both — is weak. It requires seed oils to add measurable harm on top of processing, yet controlled testing shows no oxidative-stress increase, seed oils fall rather than rise with processing, and UPF associations are already explained without invoking fat type. ³ No factorial trial isolating the four combinations exists in the claims, so H4 cannot be affirmed and is not currently needed to explain the data.

The evidence forces a narrower conclusion than either popular narrative. On controlled trials, the benefit of seed oils for blood lipids is firmly established — a 54-trial synthesis, a 'Strong' USDA grade, and convergent single-trial data all point the same way. ³ On processing, the harm of the ultra-processed matrix is equally firm: a controlled crossover trial isolating processing from composition, plus two large cohorts whose associations survive nutrient adjustment. ³ These two findings are compatible, and together they make H2 the load-bearing explanation.

The case for seed oils harming health independently of processing is not merely unproven; it runs against the weight of the CONFIRMED evidence. Its strongest single study, Sydney Diet Heart, has a confidence interval touching unity and is contradicted by far larger and more recent datasets. ² The genuine open frontier concerns interactions and long-run exposure. SPECULATIVE: because no cited factorial trial pits seed-oil-based UPFs against saturated-fat-based UPFs and against whole-food seed oils simultaneously, a residual intrinsic seed-oil effect at modern intake levels over decades cannot be formally excluded — the reasoning being that all confirmatory seed-oil benefit data are short-to-medium term and surrogate-marker based, while the hard-endpoint data are observational. ³ That is a gap in the evidence, not a positive finding of harm.

On the incentives, the picture is symmetric and worth stating plainly. The most seed-oil-favourable oxidative-stress claims come from an industry-funded body; the official benefit findings come from bodies with institutional stakes in existing guidelines; and the economic 'availability and functional properties' account originates in industry technical literature that explains prevalence without addressing health. ⁴ None of these disclosures overturns the CONFIRMED trial data, but each should temper how much any single source is allowed to settle.