The Steak Button

On 28 January 2026, Professor Christian Pfrang went to the press with a result he could have frightened people with, and refused to.

His team at the University of Birmingham had switched an air fryer on with nothing inside it, after more than seventy uses, and watched it throw out more than double the ultrafine particles it had thrown out when it was new. The cause was residue, lodged in a part of the machine no owner can reach. It was a headline, sitting there, already written. What Pfrang said instead was this: "While the effects are not producing emissions that families should be concerned about, this finding does make a case for air fryer design that allows for a deep clean to keep emissions low in the long term."⁹

The apparatus that produced that result belonged to Ruijie Tang, a PhD student in his group. She had sealed a 4.7-litre COSORI air fryer inside a Perspex chamber of 151.9 litres and cooked twelve dinners in it, over and over. Frozen chicken. Smiley hash browns. Onion rings. Broccoli, cauliflower, courgette, corn on the cob. Mushrooms, vegetarian sausages, pork sausages, unsmoked bacon, smoked bacon. A mass spectrometer sampled the air in the box once every second, and every dish was cooked at 175 °C.¹

The first thing the instruments found was an absence. On the runs where the machine was switched on empty, they recorded no nitrogen oxides at all. The NOₓ the study did detect appeared only while certain foods were cooking. The appliance itself produces none, because it has no flame — there is nothing in it to burn.¹

Scaled into a kitchen, the numbers stay small. The paper is careful that this is a scaling rather than a measurement: the chamber is roughly a hundred times smaller than a modest kitchen, so the concentrations inside it are elevated and, in the authors' words, "cannot be directly comparable to real-world conditions."¹ Modelled into a deliberately conservative 15 m³ room, peak nitrogen dioxide came out at about 1.1 µg/m³, against a WHO 24-hour guideline of 25. Modelled total volatile organic compounds came out between 0.017 and 0.24 mg/m³, against UK guidance of 0.3 mg/m³ over eight hours. The authors summarise this as "low benchmark exceedance under the scenarios considered."¹

Those two comparisons are the reassurance, and they are the whole of it. Nitrogen dioxide is measured against a WHO guideline. Volatile organics are measured against UK guidance. They are the two pollutants in this study for which anybody has published a threshold, and Birmingham's modelled kitchen sits under both.

The third pollutant has no threshold at all. Ultrafine particles made up more than 94% of every particle the study counted, and there is nothing anywhere to compare them against. The WHO, the paper notes, "sets no numeric guideline values for total VOCs or UFPs," and the EU's 2024 ambient air directive "introduces UFP monitoring, but no limits."¹

Toronto, 2023. Nanjing, 2026.

Deeper in the Results section, in its own voice, the Birmingham team writes: "air frying generates significant UFP emissions that vary markedly with food type: high-fat dishes can produce UFP concentrations comparable to deep frying, while low-fat and oiled items emit considerably lower levels."¹

The arithmetic behind that sentence is worth reading slowly. In the chamber, at 175 °C, air-fried pork sausages emitted ultrafine particles at a rate of 17.4 × 10¹² per minute — 17.4 trillion. The paper sets that against literature values of roughly 20.2 trillion a minute for pan-fried chicken with vegetables and roughly 9.6 trillion for deep-fried beef with vegetables.¹ Those two comparators are figures cited from other researchers' papers, not measurements made in this study, and they came from different rooms with different instruments. Held loosely, they still say something: air-fried pork sausages — one of the twelve dishes, and the highest-emitting of them — land at about 86% of that other study's pan-fried chicken figure, and nearly twice its deep-fried beef. On particle count, against numbers from other people's rooms, "far cleaner than frying" is not a sentence anyone has actually earned.

It gets less settled from there. In 2023, Xing Wang and Arthur Chan at the University of Toronto ran an air fryer against a pan and found the air fryer worse. What they measured was the emission factor: the particulate a method produces for a given amount of oil. Per gram of oil used, the air fryer came out 2.1 times higher than the pan for chicken wings and 5.4 times higher for chicken breast, with volatile organics 2.5 to 4.8 times higher.⁴ That is a ratio per gram of oil, not per dinner — and an air fryer uses a spray where a pan uses a pour. A higher figure per gram is perfectly compatible with far lower emissions per meal, simply because far less oil goes in. Their proposed explanation, verbatim: "higher temperature and greater turbulence lead to higher PM₁₀ EFs for cooking with the air fryer compared with the pan for the same mass of oil added." Their paper is behind a paywall, and their actual settings cannot be read by anyone without a subscription — not what temperature their air fryer ran at, not what temperature their pan ran at.

Then, in February 2026, a team at Nanjing University published a three-way comparison of a pan, an air fryer and an oven. The pan produced the highest PM₂.₅ concentrations and the highest mass-based emission rates. The oven, they report, produced minimal emissions. Then they ran their measured particles through a standard lung-deposition model, and the ordering changed: "the air fryer produced the highest particle number dose, while pan frying generated the largest mass and surface area dose."⁵ That dose is a model applied to measured particles, not a measurement of what anybody inhaled. And that paper, too, is behind a paywall.

The three are not necessarily in conflict, because they are not measuring the same quantity. A cloud of smoke can carry more particles and less mass at the same time, and cooking aerosol is exactly that kind of cloud: not soot, but recondensed droplets of vaporised oil and fat, formed when hot vapour meets cooler air. They are enormously numerous and they weigh almost nothing. Which means the entire architecture of particulate air-quality law — PM₁₀, PM₂.₅, every standard that exists — cannot see them, because every one of those standards weighs particles rather than counting them.

And notice what none of that adds up to. Of these three papers, Birmingham's is the only one that took its measurements and asked what they would mean in an actual room, with an actual air-exchange rate, against an actual guideline. Toronto published emission factors per gram of oil. Nanjing published orderings, particle diameters and a modelled lung dose. Nobody has done the room arithmetic for either of them, and nobody is obliged to: a chemistry paper is entitled to stop where its question stops. Birmingham's reassurance is bounded by Birmingham's paper. What the other two mean for the air in a kitchen has never been worked out, and none of the three papers reconciles the other two.

The steak button

The machine in Tang's chamber was a COSORI Pro LE, model CAF-L501-KUK — the 220–240 volt version, the one sold in Britain. Its manual gives a temperature range of 75 °C to 230 °C, and lists nine one-touch presets.⁶

Preset	Temperature	Time
Keep Warm	75 °C	5 min
Bacon	160 °C	8 min
Veggies	170 °C	10 min
Frozen Foods	175 °C	10 min
Seafood	175 °C	8 min
Chicken	190 °C	20 min
French Fries	195 °C	25 min
Preheat	205 °C	4 min
Steak	230 °C	6 min

Bold: beyond the range of any published emissions measurement anyone can read. Not a risk ranking — no measurement exists either way.

Now set that against the entire measured envelope of world science. The 2026 chamber study cooked everything at 175 °C.¹ The same group's earlier study, published in 2024, ran fourteen air-frying tests at three settings: 160 °C, 175 °C and 190 °C.³ Above that, the search came back empty.

Across four bibliographic databases, in four languages, screening 212 unique records and 44 forward citations: no measurement of an air fryer's emissions above 190 °C that anyone can read.

Three of the machine's own default settings sit outside that envelope. The chips preset is five degrees past the last published emissions measurement. The preheat is fifteen degrees past. The steak preset is forty.

It would not take much to close it. Put the same machine back in the same chamber with the same twelve dishes, and run it at 190, 210 and 230 °C. One afternoon. One particle counter. If the emissions come out flat, the edge is not an edge, and everything that follows here is wrong about its size.

The preheat is the strangest of the three settings, because the researchers used it. The manual tells you to preheat at 205 °C, and the Birmingham team did exactly that — every experimental day, four minutes, empty, "following the manufacturer's manual."² It is the appliance's factory default: the number on the display of one of these machines before a single chip goes in. In a paper of nearly fifteen thousand words, that number appears exactly once, in the sentence describing the warm-up. No emissions figure at that setting is reported anywhere — not in the paper, not in its tables, not in its supporting information. The study's published envelope stops at 175 °C. The published literature stops at 190 °C. The appliance's default preheat is 205 °C.

The work was paid for out of public money: a NERC grant, NE/V002414/1, now closed.²² The study answered the question it set out to answer, and it answered it at 175 °C.

The manual has nothing to say about any of this. Across it, the word "emission" appears zero times. So does "particle," and so does "air quality." The word "smoke" appears twelve times, and every one of those twelve is an instruction about the food: "Cooking greasy foods will cause oil to collect beneath the crisper plate. This oil will produce white smoke, and the basket may be hotter than usual. This is normal, and should not affect cooking."⁶ The manual for the Ninja AF100UK uses the word "emission" the same number of times.⁷ Neither manual has left anything out, because no rule asks it to say a word about what it puts into a room — not the safety standard, not the ecodesign regulations, not the general product safety rules that govern a cooking appliance sold here. This is the vocabulary the whole category was handed, and it describes a kitchen in which the only thing an aerosol can affect is dinner.

Deep-frying ran hotter and came out cleaner

The obvious next move is to say that turning the dial up must make things worse, and that the forty degrees above the last published emissions measurement are therefore forty degrees of risk. The 2024 paper appears to license this. Its abstract states that "the cooking temperature was found to be the key factor that positively correlated with both PM and VOC emission strength."³

That correlation was computed within three cooking methods, and the air fryer is not one of them. Here is what the same paper's own tables say about how the five methods compare with each other:

Method	Average cooking temperature	Peak PM₂.₅, median
Pan-frying	207.7 ± 15.7 °C	92.90 µg/m3
Deep-frying	142.5 ± 13.2 °C	7.67 µg/m3
Stir-frying	110.3 ± 11.8 °C	26.68 µg/m3
Boiling	89.4 ± 1.9 °C	0.65 µg/m3
Air-frying	— (never recorded)	0.59 µg/m3

Tang et al., 2024, Tables 1 and 2. Medians, not means.³

Deep-frying ran 32 degrees hotter than stir-frying and emitted three and a half times less fine particulate. In the 2024 study's own between-method data, temperature does not rank the methods on particle mass — the measure Birmingham's reassurance rests on. The hotter method was the cleaner one.

The reason is that there are two heat-driven mechanisms in a kitchen, not one, and the second is the bigger. The first is oil breaking down above its smoke point. Rapeseed oil's smoke point is 210 ± 11 °C — a band running from 199 to 221, and a figure the Birmingham team take from another group's work, not their own. The oil in their frying pan ran at 208 to 225 °C: it starts inside that band and finishes above the top of it. The team say so in as many words. The pan was "above the smoke point of rapeseed oil."³

The second mechanism is food pyrolysing where it presses directly against hot metal, and the paper names it repeatedly. Its own explanation for why adding more oil to a pan reduces particulate is that the oil "reduce[s] the direct contact between the food and the pan, thereby diminishing the pyrolysis of food particles."³ And the wok — the stir-frying method in the table above — never crossed any smoke point at all. Its temperatures, the team say, "were lower than the smoke points." It still emitted more than forty times what the air fryer emitted, at a median of 26.68 against 0.59 µg/m³.³

An air fryer has no seared metal surface pressed against food. It has moving air. Which means the good news is bigger than it looks: part of this appliance's cleanliness is structural. Turning the dial to 230 °C does not turn an air fryer into a frying pan, because the frying pan's dominant mechanism is not available to it.

It also means the position on those forty degrees is not the one anyone would expect. Look at what the two numbers actually are. The pan's 208–225 °C is a measured temperature of oil, taken with a handheld infrared thermometer gun pointed at the pan every minute during cooking.³ The air fryer's 230 °C is a setpoint of air, asserted by the appliance's own thermostat — and the study's supporting information confirms that nobody checked it: "The air fryer's built-in timer and thermostat controlled the heating process; therefore, no manual intervention occurred during cooking."² No thermocouple. No probe. No measurement of any temperature inside an air fryer has ever been published: not the coil, not the circulating air, not the film of oil on the surface of the food. There is one number in the public record, and it is what the dial says about itself.

Setting an air setpoint beside an oil smoke point and calling the difference a gap puts two different quantities in the same column. The smoke point belongs to the pan, where it was measured, and it stays there.

So there are two live physical possibilities, and no way to choose between them. Either the dial governs this machine's emissions the way it governs a pan's, in which case forty unmeasured degrees matter a great deal — or the air fryer's cleanliness is over-determined by its design, the pan's mechanism is simply unavailable to it, and the dial is close to irrelevant. No published measurement can adjudicate. And the strongest single piece of evidence against the temperature story is sitting in the preset table: the Bacon preset runs at 160 °C, the lowest cooking preset on the machine, and bacon was among the highest particle emitters the study found, at 13.5 trillion a minute smoked and 9.0 unsmoked, against a low of 0.1 across the twelve dishes.¹ Nobody has published the surface temperature of an air fryer's heating element at any setting, so nobody can explain that either.

Fourteen runs, three settings, and a dash

The temperature series that would settle this has never been published.

In 2024, fourteen air-frying tests were carried out across three setpoints, and no trend was extracted from them. In the table where the paper records an average cooking temperature for each method, four methods have a number and air-frying has a dash.³ The paper does not say why. The instrument that produced the other four numbers was that handheld infrared gun, pointed at an open pan — and you cannot point one of those inside a closed appliance. That is an inference, not a finding; the authors never explain the dash.

The same geometric problem appears a second time in the same paper. The team also sampled cooking particles for environmentally persistent free radicals — stable radicals that sit on a particle's surface. The sampling rig was a filter on a pump "held 15 cm above the pan."³ It sampled pan-fried chicken, pan-fried minced beef, oil heated on its own, and wipes taken from around the cooker. The air fryer is not among them, and nobody has published whether one produces them or not. The instruments of cooking-emissions science were built for an open pan on an open hob.

In 2026, the temperature was held at a single value on purpose. The supporting information gives the reason, and it is a good one: the setpoint was "fixed at 175 °C" in order "to account for differences among air fryer models and to maintain a uniform cooking temperature across the experiments."² That is a stated, reasonable methodological choice, made so that different machines could be compared with one another.

And here is the fact that makes this something other than a hole in the science. The air fryer has been studied above 190 °C — repeatedly, internationally, and with care. A 2024 open-access study cooked beef patties in an air fryer at 140, 160, 180 and 200 °C and measured benzo[a]pyrene, a carcinogen, at every step.⁸ It measured it in the beef. Of 212 air-fryer records screened, the overwhelming majority are food chemistry: acrylamide, polycyclic aromatic hydrocarbons, heterocyclic amines, aroma compounds, lipid oxidation, texture, nutrient retention — routinely at 180 to 200 °C, in journals across several continents.

The temperature axis exists. It has been run, funded, replicated and published. It runs through the plate, and it has never once run through the room. What is missing from this appliance's science is not effort or money or instruments. It is a question, and nobody was ever assigned to ask it.

Seventy uses later, they ran it empty

Nor did anyone owe us the answer. A comparative study is entitled to treat a comparator as a comparator, and Tang's group has behaved impeccably throughout. They cleaned the appliance after every single run, by hand, to the manual's standard and past it: washing the tray and rack, wiping the cavity with a damp pad, checking the heating area for debris. Then they waited until a particle sizer, a mass spectrometer and a NOₓ analyser all three agreed the chamber was back to background before the next run began.² They published open-access.

And then, after the machine had been used more than seventy times, they switched it on with nothing inside it. The ultrafine particle count in the chamber rose from 7.8 trillion per cubic metre to 18.4 trillion — more than double. In the paper's words, it was "a value exceeding the peak concentrations observed during some cooking runs."¹ Volatile organics on the empty run were up 23%. On that one device, on that one before-and-after comparison, the machine was emitting more with nothing in it than it had when new.

The residue is not in the basket. The paper locates it precisely: it collects in "the internal fan and extraction tunnel of the air fryer, which are not accessible for routine cleaning."¹ The supporting information is blunter, and fair to the design: "For safety and design reasons, the heating compartment and internal ducting of the air fryer are not fully accessible for manual cleaning."² This is what a scrupulous owner gets.

You cannot scrub your way out of it, which is exactly why the researchers asked for a different machine rather than a different cleaning routine: "Designing future air-fryer models with removable or easily accessible cooking-chamber components (i.e., fan housings, heating element shrouds, extraction tunnels) would allow thorough cleaning of all contaminated surfaces."¹ Keep it in proportion, though. It is one device, measured at two points in time, with nothing inside it. No one has published a curve past seventy uses — a British appliance kept three to five years will see many hundreds — and nobody knows whether the film reaches a steady state or keeps thickening. The University's own release states that even the degraded machine's particle emissions "were still significantly lower than those from other cooking techniques."⁹ The degradation eats into a margin. It does not reverse the finding.

What Dyson won, and what Changhyuk Kim asked for

Everything needed to answer this question already exists — in law, in a courtroom, in a preparatory study and in a British test house. None of it points at an air fryer.

The appliance is covered by a standard. EN/IEC 60335-2-9 names "portable ovens" and "roasters" explicitly in its scope, and it is a safety standard, concerned with what the heat does to your hand: shock, fire, burns, mechanical hazard.¹¹ The boundary here is not the appliance. It is the endpoint.

Britain also already has a particle-number emissions limit on a domestic electric appliance. It applies to vacuum cleaners: dust re-emission "shall be no more than 1,00 %," calculated as the ratio of the number of particles a machine blows back into the room to the number it sucked in, a limit in force since 1 September 2017.¹² Counting particles rather than weighing them is legally possible, it is already written into retained UK law, and a vacuum cleaner on sale here has to pass it. Then look at the window the regulation defines: "all dust particles of a size from 0,3 to 10 μm." Three-tenths of a micrometre is 300 nanometres, and an ultrafine particle is under 100. So even the one count-based emissions limit this country has written for a household machine begins at three times the size of the particles that dominate an air fryer's output — and no equivalent exists for any machine that cooks food at all.

Dyson supplied the other half. The company took the European Commission to court over how vacuum cleaners were being tested for their energy labels, and on 8 November 2018 the General Court of the European Union agreed and annulled the label — the tests having been run with an empty dust receptacle, rather than in the state anybody actually owns one in. The Commission, it held, was obliged to adopt a method measuring performance "in conditions as close as possible to actual conditions of use."¹³ That case was about energy, not emissions; the words "indoor air" and "particulate" appear nowhere in the judgment. The principle is settled law. Extending it from energy to emissions is a step nobody has taken.

The power to require exactly what the Birmingham team asked for exists too. The EU's Ecodesign for Sustainable Products Regulation, in force since July 2024, applies to "any physical goods." Its Annex I sets out the parameters a regulator may write requirements against. One of them is "emissions to air, water or soil released in one or more lifecycle stages of the product" — and the life cycle expressly includes "use, maintenance, repair." Another parameter covers "ease of repair and maintenance… ease of non-destructive disassembly and re-assembly." That is the researchers' fan-housing fix, written into European law two years before they asked for it.¹⁴ The regulation never once uses the word "indoor." And it is EU law. Great Britain is outside it.

The reasoning is not even foreign to the system. In 2022 the European Commission's own science service published a 492-page preparatory study on cooking appliances. On the charcoal filter inside a recirculating cooker hood, it says: "They cannot be cleaned, so need to be replaced approximately every 3 or 4 months. If the charcoal filter is not regularly changed, a significant decrease of indoor air quality can be expected." The passage is credited to APPLIA, the European home-appliance industry's own trade association.¹⁵ Residue accumulates in a part you cannot clean; performance degrades; indoor air quality falls; therefore someone must be told, and a service interval must be attached. The identical argument, four years early, aimed at a different component of the same kitchen.

And there is a laboratory. Leatherhead Food Research added air fryers to its UKAS-accredited cooking-instruction verification service in 2022, and reports that air-frying is now its second most-requested cooking technique — a figure that comes from a company press summary rather than a published methodology, and is worth holding lightly.¹⁷ The accredited lab, the accredited protocol and the appliance are already in the same room. The question that lab is accredited to answer is how long to cook the chips for.

Which? tests air fryers too, and publishes its weightings: 50% quality of cooking, 25% ease of use, 20% machine performance, 5% build quality. Emissions: 0%. Cleaning is scored — but inside "ease of use," as a chore. And in December 2025, before the Birmingham study published, Which? deepened its testing by adding more foods: frozen chips, a baguette.¹⁶ The air was never on the list.

Nobody in that sequence did anything wrong. The IEC wrote a safety standard and it is a good safety standard. Which? tests what its members ask it about. The Commission's science service scoped an energy instrument on energy, in 2014, before anybody owned one of these machines. Every one of those remits was drawn before the question existed.

Which brings us to Changhyuk Kim. In September 2025 — four months before Birmingham published — Kim and his colleagues at Pusan National University in South Korea measured ultrafine particles coming off air fryers, toasters and hairdryers, and closed their published abstract in the Journal of Hazardous Materials with a single sentence: "Regulation and test protocols for UFP emissions from home appliances are required to achieve clean indoor air quality for human health."¹⁰ A laboratory on another continent named the missing instrument, in print, in a peer-reviewed journal. Ten months on, in July 2026, no such protocol appears in any UK or EU standard, ecodesign regulation or product-safety rule.

And that paper is paywalled too. Pusan's own press release says the emissions it recorded "often varied based on the operating temperatures"²³ — which means a Korean bench may well have taken an air fryer above 190 °C and written the number down in a table nobody outside a subscription can open. If it did, the gap described here is narrower than it looks, and there is no way to find out from outside the paywall.

The extractor fan

An air fryer is a 1,500-watt convection oven with a fan. It has no flame and no combustion. Three independent teams have now put instruments in a room with one, and the only team that scaled its results into a kitchen found nitrogen dioxide and volatile organics below the thresholds anyone has published for them. The scientist who found the residue result went straight to the press and told families not to worry, and his own institution confirmed that even a degraded machine beats a frying pan. The appliance has been cooked in at 200 °C hundreds of times by careful people with good instruments, and every one of them measured the thing that actually enters a human body: the food.

Meanwhile the gas hob stands three feet away, burning methane in an open room. CLASP's field study across seven European countries found around twice the level of nitrogen dioxide in homes cooking with gas as in homes cooking with electricity. Over half the gas-cooking households exceeded WHO daily guideline values.¹⁹ A public body with finite money has, arguably, spent it in precisely the right place. Take forty degrees of unmeasured dial on a machine whose modelled kitchen sits under every threshold anyone has published for it, build from it an indictment of an entire regulatory architecture, and you have found the edge of a graph and mistaken it for a cliff.

Every word of that is true. And not one word of it says what the machine does at 230 °C.

The reading this evidence invites — that a machine bought for being clean turns out to be dirty — is not what any of it shows. Nothing here demonstrates a harm to anybody. Emitted is not inhaled; inhaled is not absorbed; absorbed is not harmful; and the distance between those four words is where most kitchen scares are manufactured. What the evidence shows is narrower and stranger. Three peer-reviewed teams have measured an appliance that an industry survey of 2,057 UK adults put in 61% of British households¹⁷, they measured different quantities in different rooms on different metrics, a British citizen cannot read two of their three papers without a subscription — and the settled scientific method for answering the question has never once been pointed at the room.

Which is why the one action worth taking here is not about air fryers at all.

The Birmingham paper models the same emissions into a 30 m³ kitchen with a ducted extractor running, and gets peak ultrafine particle levels "more than 10 times lower" than in the unventilated case.¹ It is the only lever in this story with a measured number behind it — bigger than the residue effect, bigger than any plausible effect of the dial. What it measures is not capture at source but the air of the room as a whole: a ducted hood pulls kitchen air outside and lets clean air in behind it, so everything suspended in that air comes down, whether or not it ever passed under the hood. This is why an extractor still works on an air fryer sitting on the worktop, outside the hood's capture zone entirely.

A recirculating hood does none of it. As the Commission's own study puts it, its filters exist "for filtration of odours before air is returned to the room."¹⁵ And you cannot assume yours is the useful kind: in a national survey of 1,861 British homes, 71% did not have the ventilation provision that building standards require — standards whose demands include mechanical extraction in the kitchen.¹⁸

So: run the extractor when you cook, if it vents outside; open a window if it doesn't. It costs nothing, and it works for every method of cooking in the room.

The dial is a map rather than a recommendation. Cooking at 180 °C with a spray of oil puts the machine inside the envelope where every emissions measurement anyone can read was taken. Inside that envelope, its particulate mass — the metric that cannot see most of what this machine emits — straddles the room's own background, at a median peak of 0.59 µg/m³ against a background of 0.5.³ Its particle number — the metric that can see them — is where the three teams disagree. The Steak preset is forty degrees outside that envelope, where neither number exists at all. That is the whole of what anybody can tell you, and it is enough to choose by.

It is not a warning, and the symmetry has to be said out loud: 230 °C might well turn out to be cleaner. The Steak preset runs for six minutes, the shortest cook on the machine; the chips preset runs for twenty-five, the longest. Every emission figure in this evidence is a rate per minute, and what a room actually receives is that rate multiplied by the minutes the machine runs. So turning the dial down can mean cooking four times as long, and a cooler, slower, longer cook might put more into the air in total than a hot fast one — or less. Nobody has published that calculation, and nobody can, because the rate at 230 °C does not exist.

Cleaning the basket is not the emissions control, because the residue is not in the basket. It is in the fan housing and the extraction tunnel, and you cannot reach them. That is not a failure of effort on anyone's part. It is the reason the researchers asked for a redesign.

Which makes the shopping advice unusually short, because there is almost nothing to buy. Of the air fryers on sale in Britain, none reviewed for this report offers a removable fan housing or an accessible heater shroud, and only one — the Instant Vortex Plus — has a filter in its exhaust path. What Instant Brands claims for it is odour: "Built in OdorErase air filters remove cooking odors."²¹ No test of it against ultrafine particles has ever been published — not by the manufacturer, not by Which?, not by the Birmingham team, who tested a COSORI, which has no filter at all. So it is not a recommendation. One machine on the market with a fan housing that came out for cleaning would settle the design argument outright.

And the Nanjing study offers one last comparison, with both its halves attached: of the three appliances it ran, the conventional oven produced the least by mass — and the air fryer the highest modelled number dose. The disagreement does not go away when you change machine.

Fedde van der Weij, June 2006

The odd thing is that the filter was imagined at the very beginning.

In June 2006 a Dutch inventor named Fedde van der Weij filed the patent that became the air fryer; it was granted in 2012. Across its description and its claims the word "health" appears zero times, and "calorie" zero times. The word "fat" appears exactly once, and it is not about the fat in anybody's body: "These openings can be provided with a filter to reduce fumes and exhaust of fat particles."²⁰ It is a line about kitchen smells and grease — a nuisance, not a hazard, and there is nothing in the document to suggest the inventor meant anything more by it. It never entered the claims.

But the physical fact that this machine blows fat particles into a room was written down at the moment of its invention — fewer of them, by weight, than the frying pan in Birmingham's own table put out, but written down all the same. Twenty years later, in a Perspex box in Birmingham, Ruijie Tang measured those particles and found them collecting in the one part of the machine nobody can open.

The button is still on the front of it. It says STEAK. Press it and the machine runs at 230 °C — forty degrees past the last emissions measurement anyone can read, and for six minutes, against twenty-five for a tray of chips at 195 °C.

Whether that is worse, or better, or exactly the same, nobody can tell you. Nobody has been asked to look.