A curious thing has happened to cooking oil. Canola, sunflower, soybean and other "seed oils" — pressed and refined from plant seeds and rich in the omega-6 fat linoleic acid — have become the target of a popular campaign claiming they quietly poison the modern diet. The charge is that these oils oxidize into toxic byproducts when heated and flood the body with inflammation-driving omega-6 fats.

Against that popular verdict stands a large and mostly consistent body of nutrition science. Peer-reviewed feeding trials and meta-analyses report that linoleic acid lowers cholesterol and cardiovascular risk when it replaces saturated fat. ² The American Heart Association explicitly endorses omega-6 fats as part of a healthy diet. ¹

The complication is that seed oils rarely travel alone. They are staple ingredients in chips, packaged baked goods and fast food — the "ultra-processed" products that a separate and growing literature ties to obesity, heart disease and early death. ² The scientific question is therefore an attribution problem: are seed oils themselves harmful, or are they simply riding along inside foods that are harmful for other reasons? This report separates what the evidence shows about the oils from what it shows about their company.

The most direct way to judge seed oils is to test them as a single variable, and here the controlled evidence is unusually clean. A meta-analysis of metabolic-ward studies — the most tightly controlled feeding setting, where every calorie is supervised — found that swapping 5% of dietary energy from saturated fat or carbohydrate for omega-6 polyunsaturated fat cut LDL cholesterol by 0.39 mmol/L, or about 15 mg/dL. ¹ That is a clinically meaningful drop from a simple ingredient substitution.

The benefit carries through to hard outcomes. A pooled analysis of eight randomized trials with 13,614 participants found that increasing polyunsaturated fat in place of saturated fat lowered coronary heart disease events by about a fifth (relative risk 0.81, a 19% reduction). ¹ In diabetic and dyslipidemic patients, a systematic review of 11 trials reported that oils from canola, flaxseed and sesame improved both blood-fat profiles and blood-sugar control. ¹ One trial in that review gave 46 adults with type 2 diabetes 30 ml of sesame oil daily for 90 days and saw significantly lower blood glucose and lipid-peroxidation markers than a soybean-oil comparator, alongside higher antioxidant-enzyme activity. ¹

A caution belongs here on comparing doses. These studies use different yardsticks — some measure grams or millilitres, others a percentage of dietary energy, others blood biomarkers — so their effects are not directly interchangeable without knowing each group's total calorie intake. ³ What they share is design: each holds the food context steady and moves only the oil, and each points the same direction.

Open: No trial in this evidence base compared whole foods cooked in seed oil against identical foods cooked in butter or tallow while holding processing level constant.

If seed oils behave one way in a research kitchen and another way inside a bag of chips, that distinction should show up in the evidence. It does not — not because the answer is settled, but because no study has looked directly.

The available evidence addresses processing context only at arm's length. An umbrella review pooling 48 studies and 206 meta-analyses found that canola, virgin olive and rice bran oils significantly lowered total cholesterol compared with other vegetable oils, a reduction ranging from 0.86 down to 0.11 mmol/L across oils and comparators. ¹ Harvard's systematic review of prospective cohorts found that consuming linoleic acid at 5–10% of dietary energy was associated with lower coronary heart disease risk. ¹ Both describe outcomes at particular intake levels; neither separates oil eaten in a home-cooked meal from the same oil delivered inside a packaged snack.

That gap is the crux of the whole debate. The two cholesterol figures above are not strictly comparable — one isolates a controlled replacement protocol, the other aggregates many oils and comparators — and neither was designed to test whether the food matrix changes the verdict. ² The honest summary is that the beneficial signal for seed oils comes overwhelmingly from minimally processed or research-diet contexts, and the literature simply has not tested them where critics are most worried: inside ultra-processed food.

Open: No cohort or trial has stratified seed-oil intake by NOVA processing level to report disease outcomes separately for oils in minimally processed versus ultra-processed foods.

Population studies following tens of thousands of people offer a different vantage: not what happens under supervision, but what happens over years of ordinary eating. Here the linoleic-acid signal is favorable and reasonably consistent. An NHANES analysis of 49,884 US adults, with 7,347 deaths, found that those in the highest quartile of linoleic acid intake had 13% lower overall mortality (hazard ratio 0.87). ¹ The same analysis found that replacing 5% of energy from saturated fat with polyunsaturated fat was linked to 12% lower mortality. ¹ A 2020 meta-analysis of 44 cohorts covering more than 800,000 people is reported to have found the same 13% mortality reduction, though that figure reaches us through a single interested-leaning site and should be read as reported rather than confirmed. ¹

The cohorts converge, but the trials do not follow them all the way. The Cochrane Collaboration's review of 49 randomized trials and 24,272 participants — the gold standard for this kind of question — concluded that increasing polyunsaturated fat probably has little or no effect on all-cause mortality (relative risk 0.98, its confidence interval 0.89–1.07 crossing 1.0). ¹ This is a genuine divergence in point estimates and statistical significance rather than incompatible ranges: the Cochrane confidence interval (0.89–1.07) overlaps the cohort point estimates (HR 0.87–0.88), and it has a plausible explanation rather than a contradiction: cohorts capture decades of habitual eating and substitution effects, while trials are shorter and may lack the duration or statistical power to move a mortality endpoint. ¹ On coronary events specifically the two designs agree more closely, with independent syntheses reporting CHD event reductions of 15–19% and an associated 21% CHD-death reduction. ²

Crucially, none of these cohorts separates seed oils eaten in salads from seed oils eaten in fried and packaged food. They adjust for nutrients, not for processing level, so they cannot by themselves resolve the attribution question.

Open: The 62% higher mortality with heavy ultra-processed intake rests on a single cohort with 335 deaths and no independent replication of that magnitude [C-026].

The anti-seed-oil case rests on two mechanistic claims — that linoleic acid inflames the body and that it oxidizes into dangerous compounds. When researchers have gone looking for those effects in humans, they have largely come up empty.

On inflammation, four independent lines of evidence agree. A systematic review of 15 randomized trials found no significant effect of dietary linoleic acid on CRP, TNF-α, fibrinogen or other inflammatory markers, though this reaches us via a single opinion site and is graded reported. ¹ A controlled crossover trial in 17 adults at cardiometabolic risk found no difference in six inflammatory biomarkers between a canola-oil control snack and its comparator. ¹ The 48-study umbrella review found no change in CRP for canola, flaxseed, olive or coconut oil. ¹ And Harvard's review of randomized feeding studies found dietary linoleic acid did not raise inflammatory markers. ¹ Different designs, same null result.

The fear of an omega-6 "overload" leans on the idea that linoleic acid converts into inflammatory arachidonic acid. In humans it barely does: that conversion runs at less than 1%. ¹ A 12-week double-blind trial that swung linoleic acid intake from 2.5% to 10% of energy found arachidonic acid concentrations did not differ between the arms, suggesting even modest intakes already sustain near-maximal synthesis. ¹ A 2011 review reportedly found no impact on arachidonic acid even with a 90% cut or 600% increase in intake, though this also reaches us via a single opinion site and is graded reported. ¹

The genuinely inconvenient facts for seed-oil defenders sit in the same trial. High linoleic acid intake did markedly suppress plasma EPA, a beneficial omega-3, and raised the arachidonic-acid-to-EPA ratio. ¹ And in the test tube, linoleic acid oxidizes roughly 100 times faster than saturated stearic acid at 100°C. ¹ But those are mechanistic signals, not clinical outcomes — and the oxidation ratio is a laboratory maximum that does not directly translate to real cooking temperatures or foods with their own antioxidants. ¹ Against them stands a wall of human-outcome data, including a 29% coronary reduction reported by an industry-aligned source and a 15% event and 21% death reduction across 13 cohorts of 310,602 people. ²

Open: Whether repeatedly consuming oxidized-lipid byproducts from cooked seed oils over decades carries clinical risk has not been tested in any trial in this evidence base [C-029].

When major bodies and meta-analyses are asked to apportion blame, they point away from the oils and toward the way food is made. The evidence that ultra-processed foods harm health independently of their nutrients is strong and comes from both experiments and observation.

The experimental cornerstone is the Hall inpatient trial: when an ultra-processed diet was matched to an unprocessed one for calories, sugar, sodium, fiber and macronutrients, participants still ate about 508 extra calories a day and gained weight. ² Something about processing itself — not any single ingredient — drove the overeating. Cohorts echo this. In the NutriNet-Santé study of 105,159 adults, each 10% rise in ultra-processed intake was tied to a 12% higher cardiovascular risk that survived adjustment for saturated fat, sodium and fiber. ¹ In Spain's SUN cohort, heavy ultra-processed consumers had 62% higher mortality (a single-cohort finding with a wide confidence interval 1.13–2.33 and only 335 deaths), an association not attenuated by adjusting for saturated and trans fats, added sugars and sodium. ¹ Across 17 cohorts, 93% of ultra-processed–obesity associations held up after nutrient adjustment. ¹

Seed oils are, by definition, on the other side of this line. The NOVA classification places vegetable oils in the processed culinary ingredients group, not the ultra-processed category. ² Dr. Christopher Gardner of Stanford frames the practical consequence: the real concern should be overeating ultra-processed foods and their added sugars, sodium and high-fructose corn syrup, not the seed oils inside them. ² He goes further as a matter of scientific caution — arguing, in a view graded speculative, that until a trial compares ultra-processed food with and without seed oils, any health change reflects the combination of all ingredients, not the oils alone. ¹ The review authors behind one journal analysis likewise suggest, speculatively, that ultra-processed harm may stem from additives, packaging and neo-formed contaminants rather than traditional nutrients. ¹ And Cochrane, with its moderate-quality rating and wide confidence intervals, offers institutional endorsement tempered by acknowledged uncertainty. ¹

Open: The specific 29% coronary reduction and the one-third diabetes risk reduction attributed to linoleic acid both come solely from an industry-aligned source and lack independent corroboration [C-016][C-017].

Four competing explanations can be laid over the same evidence, and they are not equally supported.

The first holds that seed oils are independently harmful. Its entire affirmative case is two mechanistic findings: suppressed EPA at high intakes and rapid oxidation in the laboratory. ² Arrayed against it is nearly every human-outcome study in the file — the null inflammation results, the sub-1% arachidonic-acid conversion, the cholesterol and coronary benefits, and the mortality associations. ⁴ This hypothesis stands weak: it privileges mechanism over measured clinical effect.

The second reading is the mirror image — that seed oils are beneficial or neutral. It draws on the whole weight of the outcome literature and is contradicted only by the same two mechanistic signals. ⁴ It is the best-supported claim about the oils themselves. Its limit is that most supporting trials are shorter-term and use mixed food contexts, so it cannot claim to have tested oils inside ultra-processed products.

The third explanation concerns the other variable entirely: ultra-processed foods harm health regardless of seed-oil content. Nothing in the evidence contradicts it, and much supports it — the macronutrient-matched overeating in the Hall trial, the nutrient-adjusted cohort risks, the persistence of obesity associations after adjustment. ³ This hypothesis is well supported.

The fourth, and most consequential, is that the harm blamed on seed oils is confounded by the ultra-processed matrix they inhabit — neutral or beneficial in home cooking, guilty only by association in junk food. ⁴ It is plausible and coheres with everything above, but it remains untested, because no cohort stratifies oil intake by processing level and no trial isolates the oil within an ultra-processed product. ¹ All four hypotheses converge on the same missing experiment: an ultra-processed diet formulated with seed oils versus one formulated with saturated fats, everything else held constant. ³

On the narrow question of the oils themselves, the evidence is strong enough to force a conclusion. Studied as a single variable, seed oils rich in linoleic acid lower LDL cholesterol, reduce coronary heart disease events by roughly 15–19% when they replace saturated fat, and do not raise inflammation or arachidonic acid in humans. ⁴ The popular claim that seed oils are inherently harmful is not supported by the human-outcome evidence assembled here; its case rests on mechanistic signals that clinical measurements have repeatedly failed to confirm. ³

On ultra-processed foods, the evidence is likewise firm: they raise disease and death risk beyond what their nutrients explain, and they drive overeating even when macronutrients are matched. ³ That harm is real and largely independent of any single ingredient.

Where the evidence stops short is the bridge between these two findings. It is a documented fact — not speculation — that no study has isolated seed oils inside an ultra-processed food, and that seed oils are classified as culinary ingredients rather than ultra-processed products. ³ The pattern is consistent with seed oils being a bystander in ultra-processed harm rather than its cause, but that remains an inference from the available evidence, not a demonstrated result, because the disaggregation trial has never been performed. Two grades of caution apply: several favorable figures rest on single interested-party sources, and the mortality benefit itself is contested between converging cohorts and a null Cochrane trial synthesis. ⁴ The debate will not be closed by argument. It will be closed by the disaggregation trial that Gardner and the hypothesis structure both identify as the missing piece. ³