The short version

  • Rank food by suffering per gram of protein rather than by whether it comes from an animal.
  • Out: chicken and duck, eggs, farmed fish, prawns and other decapods, squid and octopus. No welfare label in Australia rescues any of them.
  • In, freely: milk, yoghurt, whey, cheese, oysters and mussels, kangaroo, labelled-wild venison, and RSPCA Certified free-range pork.
  • In, occasionally: grass-fed beef and lamb, rangeland goat, named large wild-caught fish. Unlabelled pork in social settings only.
  • Pork belongs on the same scale as everything else, contrary to the first version of this post. A breeding sow’s farrowing crate contributes about 2% of pork’s confinement-days per kilogram; the ordinary indoor life of the pigs eaten contributes most of the rest. Free-range plus RSPCA Certified is enough to move pork into acceptable, and nationally available at Coles.
  • Dairy’s welfare cost is lameness in the milking cow, not the bobby calf. Between 4% and 19% of an Australian dairy cow’s life is spent on a painful foot, and the two big local surveys disagree by a factor of five.
  • Whey is not environmentally free. It now earns about 30% of the revenue from a litre of cheese milk, so it carries about 30% of the footprint.
  • Cheese costs under three times tofu per gram of protein, not four. The standard dataset hands cheese the entire burden of its milk.
  • The passes for kangaroo, beef and lamb are Australia-specific and do not travel.

The vegan/omnivore binary doesn’t really track what I care about. I want to know how many days of confinement and pain a gram of protein requires, and animal foods differ by up to a thousand times on that measure. Removing the worst categories captures most of the ethical benefit of veganism for a fraction of the effort and inconvenience. Dairy mostly stays, chicken goes, and the best-scoring meat in Australia is kangaroo.

Brian Tomasik’s estimates supply the metric. He counts per kilogram of food, which flatters milk because it is mostly water, so I divide by protein content, as he says he would if he rewrote the piece. Small animals produce little meat per life, so chicken and eggs embed far more confinement-days in each gram of protein than beef. His inputs are American, so I rebuilt milk, beef and eggs with Australian ones: a pasture dairy cow’s herd life and yield, her ordinary days scored like a grazing beef animal’s rather than his housed American cow’s, and her lameness and calving charged on top; cattle that live two to three years before slaughter rather than thirteen months; and supermarket eggs, only about 13% of which are cage-laid, with cage-free days counted at half the pain. On that basis chicken causes about 70 times the suffering of milk per gram of protein, eggs about 105 times and beef 7 to 10 times. Re-running his calculator with Rethink Priorities’ sentience weights, as Bentham’s Bulldog did, widens the chicken gap to about 150 times. I give suffering roughly three quarters of the decision and environmental harm one quarter. That split is my judgement rather than a research finding, and it moves ruminant meat into the occasional category while changing little else. A heavier environmental weight would push beef and lamb lower down.

What I stopped eating

Chicken and duck, including RSPCA Certified. Around 99% of Australian growers use the fast-growing Cobb 500 and Ross 308 breeds. The RSPCA standard regulates stocking density and litter but leaves breed untouched, even though the Welfare Footprint Project attributes about two thirds of a broiler’s disabling pain to breed alone. The label fixes the shed and leaves the bird.

Eggs, even pasture-raised. Cage-free systems roughly halve waking hours in pain compared with cages, which matters, but eggs remain a top-tier harm per gram of protein and are several times worse than beef per meal. Australia kills around 12 million male chicks a year at hatching, and I want to be consistent about what that costs here. Hatcheries either macerate, which the RSPCA describes as instantaneous, or gas with carbon dioxide or argon, which takes 10 to 15 seconds to induce unconsciousness. Fifteen seconds per chick, set against the hen’s own lifetime of pain-hours, rounds to nothing on this metric. The case against eggs is the hen, not the chick.

Male chick culling is ending regardless, which matters to readers who do count deaths. Specialised Breeders Australia installed the first in-ovo sexing machine in the Asia-Pacific at its Victorian hatchery, with sexed chicks available from 1 June 2026 and the first eggs from those birds due in shops around the turn of 2027. Europe is several years ahead, with in-ovo sexing above a quarter of the EU market and culling banned in Germany and France.

Rearing the males, as dual-purpose flocks did before laying and meat lines diverged, fares badly on the same metric: layer cockerels take 8.5 to 18 weeks and about six kilograms of feed to produce a 1.3 to 1.5 kilogram bird. Technology is likely to retire the culling objection, while the confinement cost per gram of protein may still be standing in 10 years.

A backyard flock is the exception because the confinement argument mostly dissolves. A hen foraging at low density is closer to the zero-confinement logic that makes kangaroo rank well, leaving provenance as the main question. Point-of-lay pullets from a commercial hatchery carry their rearing-shed weeks upstream. Rescued ex-commercial hens are the clean case: the marginal egg costs nothing extra and the bird’s life improves. A neighbour’s or market’s “free-range” eggs do not inherit this exception unless the birds are traceable, which they usually are not.

Farmed fish, prawns and other decapods, and cephalopods. Whole-of-life confinement keeps farmed fish high-harm per gram of protein even when slaughter is comparatively good; Tasmanian salmon is percussively stunned, which is better than most. Prawns usually die by slow asphyxiation, and about half the prawns consumed in Australia come from regions with worse practices. Squid and octopus have high neural complexity and no welfare standards. On a menu I assume that salmon and unspecified fish are farmed.

What I eat freely

Dairy. Milk, yoghurt and whey sit near the bottom of the suffering table per gram of protein, roughly 70 to 150 times below chicken. My first version pointed at more than 300,000 bobby calves slaughtered during their first week each year in Australia and said dairy would cost more if killing carried weight. I do not weight killing itself. My concern is suffering during life, including fear and pain during death, so a genuinely instant death costs close to nothing on this metric. The calves’ first week counts; the fact of slaughter does not.

The larger charge belongs to the lactating cow through lameness. Two large Australian surveys using similar herd numbers and pasture systems disagree by a factor of five, with 3.8% of cows lame at any moment in one and 19.1% in the other. Different scoring scales do not obviously explain the gap. Point prevalence is a share of days, so somewhere between 4% and 19% of an Australian dairy cow’s life is spent on a painful foot. Episodes last for weeks and hoof lesions can take up to 35 days to resolve.

Much of it goes unmanaged. Farmers in those surveys identified only 0.82% and 5% of their cows as lame, and nearly three quarters of Australian dairy farmers keep no lameness records. Clinical mastitis is real but about an order of magnitude smaller, at roughly one painful day per lactation. Dairy still beats chicken and eggs by well over an order of magnitude, though by less than “near-zero” suggested. The honest asterisk belongs to the cow rather than the calves.

A reader asked about the cow’s pregnancies, her calvings and the removal of each calf, which I hadn’t assessed. Australian dairies separate calves within 24 hours of birth, and every controlled study finds that the earlier the separation, the smaller the cow’s response: she calls and searches for one to three days, but none of the studies I read measures distress in the mother beyond 72 hours. Calving disrupts lying on the day before and the day itself; 9% of Australian calvings across 1.7 million records were recorded as difficult births, and milk fever, retained placenta and metritis each strike between 3% and 10% of cows.

Pregnancy is also why she is dried off, often abruptly, about two months before calving. Her udder engorges for a few days. Among cows still giving 16 kg or more a day, one in six flinched, vocalised or lifted a leg at udder palpation at least once in the first three days, and high yielders spend less of the first day lying down, a standard sign of discomfort in cattle; in cows giving under 15 kg, the stress-hormone response is small. Averaged across the herd, that adds less than one distressed day.

Together, separation, calving and dry-off cost roughly three to eight distressed days per lactation, which is about 1% to 2% of the cow’s time against the 4% to 19% she spends lame. Pregnancy itself is the term I cannot price: no study quantifies gestation as a welfare cost, and the evidence that exists is confined to the final days, which the calving figure already covers. Charging separation, calving and dry-off raises dairy’s load by between a twentieth and a half, depending on which lameness survey is right, and leaves the ranking untouched.

I had also treated whey as environmentally near-free because it is a cheese byproduct. That argument depended on price rather than physics, and the price moved: whey protein concentrate roughly doubled in the year to early 2026 as demand for protein from weight-loss drugs rose, while cheese fell. On my arithmetic, whey now earns about 30% of the revenue from a litre of cheese milk. Under the same allocation convention that once made a cheap byproduct “free,” it must carry about 30% of the milk’s footprint. Concentrating that burden into a small mass of powder puts whey above cheese per gram of protein rather than below it.

The correction improves cheese. The dataset behind my old four-times figure lists every co-product allocation its authors performed, yet “whey” appears nowhere in its 76 pages. Cheese was carrying the full footprint of the milk. Allocate 30% to whey and cheese falls from roughly 24 kg of CO2 per kilogram to about 18, leaving it under three times tofu per gram of protein. I still pay that premium because halloumi, paneer and other fresh cheeses are the only dairy proteins I can cook like meat, browning rather than melting. The premium is smaller than I had claimed, and whey was never free.

Bivalves. Oysters and mussels are most likely non-sentient, and Australian longline farming is environmentally net-positive. They are the rare animal foods I treat as plausibly better than nothing.

Kangaroo is the best-scoring meat I can buy here on both axes. The animal lives wild without confinement, transport or an abattoir; each death yields a large carcass, methane is negligible and no land is dedicated to production. Commercial shooters are licensed, and an RSPCA figure puts headshots at about 96%, though the underlying data are from 2002.

Dependent young provide the asterisk. The code requires pouch young of a shot female to be killed by blunt trauma, which costs little on my metric if death is as immediate as intended. Young-at-foot can escape and starve, which counts in full. Male-only product lines largely remove both problems, making the label worth checking. Quota-managed populations are culled regardless, so commercial harvest competes against a worse culling counterfactual rather than against no death.

Wild venison follows the same logic. Deer are a declared invasive pest and are culled regardless. Field shooting operates under a national standard, and removal benefits habitat, so the usual ruminant methane objection collapses. Provenance is the catch: only labelled wild venison counts, since an unlabelled product may be farmed.

Pork, with two labels. I needed two attempts to get this category right. Per gram of protein, pork sits near beef and well below chicken because one pig produces a large amount of meat. On the environmental quarter it is the cheapest meat after chicken, at about 12 kg CO2e/kg.

I first argued that these figures missed suffering concentrated in the breeding sow. They do not; sow suffering needs a denominator like anything else. A sow spends four to five weeks in a farrowing crate for each of 2.3 litters a year and produces 2,276 kg of liveweight annually. That is about 0.03 crate-days per kilogram, compared with 1.4 confinement-days per kilogram from the pigs eventually eaten. The crate contributes about 2% of the total, or roughly 10% to 20% if one crate-day is weighted five or 10 times more heavily for severity.

Farrowing crates remain the most severe legal confinement in Australian farming, at as little as 0.5 metres wide and 2 metres long under the national code. They are not what dominates the per-kilogram figure. The ordinary indoor life of the pigs on the shelf does: roughly 90% spend about 160 days inside.

Sentience does not distinguish pork from beef. Rethink Priorities puts pigs at roughly half a human’s welfare range and chickens at about a third, while never assessing cattle; it expects cattle to land near pigs rather than below them. The difference in Australia is the life. Beef animals spend most of it on pasture, while around 90% of pigs do not.

Gestation stalls have largely disappeared and sows are now loose-housed for about 105 days of pregnancy. Investigations nevertheless found stalls in use well after the industry’s 2017 phase-out date. That phase-out remains voluntary almost everywhere, with only the ACT and Tasmania legislating. The ACT banned sow stalls in 2014 but has no intensive piggeries, so the ban costs nothing there. Tasmania’s Animal Welfare (Pigs) Regulations 2025 prohibit sow stalls and cap farrowing-crate confinement at 42 days per gestational cycle. This is the first hard legal ceiling on farrowing crates in Australia.

Pork also has the worst labelling in this guide. “Sow stall free” still allows farrowing crates, and “outdoor bred” means a pig was born outside before being finished in a shed. Yet the RSPCA label works better here than it does on chicken. RSPCA Certified pork prohibits farrowing crates, sow stalls, teeth clipping, tail docking and surgical castration, and audits the abattoir as well as the farm. It reaches the central harms rather than regulating around them.

I assumed availability would be the catch because only about 5% of Australian sows are free range and the certified share is smaller again. Those base rates do not describe what I can buy; I should have checked the shelf rather than the statistic. Coles own-brand fresh pork is both free-range and RSPCA Certified, comes from the Great Southern region of Western Australia, and is the only RSPCA-certified fresh pork stocked nationally by a major Australian supermarket. The RSPCA page says its pigs “are never kept in cages or crates of any kind”. Woolworths own-brand is only sow-stall-free, which leaves farrowing crates untouched.

Pork therefore enters my kitchen only when the label carries both properties. RSPCA Certified removes the crates and mutilations; free-range removes the indoor life that accounts for nearly all the harm in the arithmetic.

Slaughter remains. The RSPCA standard permits electrical and gas stunning, while more than 85% of pigs processed in Australia are stunned with CO2. High-concentration CO2 takes 21 to 60 seconds to induce unconsciousness, and EFSA finds that period painful, frightening and distressing to breathe through. The Western Australian processor is export-accredited, and every export-accredited pig abattoir in Australia uses CO2, so I assume my pork is gas-stunned. I accept about 45 seconds of this against a 160-day life on the same terms as kangaroo joeys, and it remains smaller in expectation than the shot-wounding rate I already accept for kangaroo.

What I eat occasionally

Beef and lamb are the lowest-suffering farmed meats. Only about 4% of the Australian herd is in a feedlot at any moment, though roughly 45% of slaughter is grain-finished for the last ~105 days. Labelled grass-fed is therefore the better purchase. Rangeland goat, which makes up most Australian goat meat, belongs in the same tier: a wild pasture life followed by conventional mustering and transport.

Unlabelled pork stays confined to social settings, where I am not choosing the label. It sits last in this ordering rather than outside it because its harm per gram of protein remains near beef. Treating it like chicken would move away from the metric I am using; the reasoning sits with the labelled pork above.

None of the Australian passes travels. Feedlot norms, harvest codes and labels change overseas, where I use a cruder rule: prefer the largest animals, avoid dishes built from small animals, and treat dairy and plants as safe.

Change log

  • 23 July 2026: added pork. Corrected the Rethink Priorities claim (its welfare ranges do not cover cattle) and re-sourced two Australian pig-farming figures.
  • 26 July 2026: rewrote dairy. Whey is not environmentally free, dairy’s welfare cost is the lactating cow rather than the bobby calf, and cheese’s footprint falls by about a quarter. Rewrote pork; RSPCA Certified free-range pork, kangaroo and wild venison moved into the freely consumed list.
  • 22 September 2026: priced the dairy cow’s separation, calving and pregnancy after a reader asked; ranking unchanged. Eggs section no longer charges chick culling as a death.
  • 26 October 2026: switched the metric from suffering per kilogram of food to per gram of protein throughout, which narrows dairy’s lead over chicken from the hundreds to roughly 70 to 150 times; rebuilt Tomasik’s milk, beef and egg figures with Australian inputs, which puts beef at 7 to 10 times milk and eggs at about 105; the 800-times figure was per kilogram and came from Bentham’s Bulldog, not Rethink Priorities. Added dry-off udder pain to the dairy cow’s costs and a chart. Rankings unchanged.

Sources