Evidence & sources

We have not cooked any of this ourselves

AirfryerHQ has run no controlled experiments. Nobody here has cooked a batch of chips, timed it, measured a core temperature and published the result. Not once, and the number of experiments behind this site is zero rather than small.

Every figure on this site therefore comes from one of three places, and each page says which: a manufacturer's own documentation for a specific appliance, a national food-safety authority, or an AirfryerHQ estimate reasoned from those and labelled preliminary. An estimate is never presented as a measurement, and no figure is ever derived from a neighbouring one — a frozen time is not a fresh time with minutes added, and a reheat is not a cook.

This is the product's actual position rather than an apology for it. A site that publishes a manufacturer's tested figure and says so is more useful than one that publishes an unsourced number in a confident voice. Where we hold nothing, we say so and refuse, which is why some states on some pages carry no time at all.

When real measurement starts, it will appear as its own kind of evidence with its own confidence rating, and this section will say so instead.

Three separate questions

Where did this come from, how sure are we, and who does it apply to. They are recorded separately because they routinely disagree. A manufacturer's cooking chart is authoritative provenance, high confidence, and applies to one model — publishing its numbers as general advice is the most common error in air-fryer content.

There is no single trust score anywhere in this product and none may be added, because no one number can express all three.

Confidence, in words

High: supported by authoritative or independently tested sources and consistent across them. Moderate: well supported, from a narrower base than we would like. Preliminary: a starting point rather than a settled answer. Disputed: credible sources give materially different answers, and we publish both positions with the likely reason for the difference.

Every figure carries its confidence word on the page where it appears, next to the number rather than in a footnote. Preliminary figures were nearly all of them when this site began and are a minority now, as manufacturer documentation was read and hypotheses were retired against it — several of them because the published figure turned out to disagree with what we had estimated. Each preliminary figure that remains still renders its limitations and its governing safety rule above the number.

Why some sources are linked and six are not

A source here carries a link only if somebody opened the document and recorded the date they did it. That is a rule in the schema rather than a habit: a source cannot store a URL without also storing the date it was retrieved, so a link on this site cannot be a guess with a hyperlink on it. A plausible-looking address is precisely the detail that makes a wrong citation look right.

Of the 61 sources in the register, 55 were retrieved and read, and every one of those 55 is linked wherever it is cited. The remaining six are not linked, because there is nothing honest to point at: five are AirfryerHQ's own reasoning, which is not a document, and the sixth is the United States Department of Agriculture's safe-minimum-temperature page, which has refused four automated requests across three sprints. We describe what it says and we do not link a page we could not open.

That split is exact and it is not a coincidence. The 55 we link are the 55 marked passage-verified — somebody read the specific passage relied on, and the reading is recorded. The six we cannot link are the six marked citation-checked, which means the work exists and is described correctly but nobody in this project has read the passage. A source's status and whether it has a link are the same fact told twice.

Where a link goes somewhere unexpected, we say so on the citation. Most manufacturer manuals here were not retrieved from the manufacturer — a Ninja booklet fetched from Argos's manual library, for instance, because Ninja's own support portal redirects to an authentication gate. Two were re-retrieved from the manufacturers themselves in August 2026, from Groupe SEB's and Russell Hobbs's own asset domains, and both turned out to be longer documents than the retailer copies of the same manuals: one carried a control panel the shorter copy omitted entirely. Where that has happened we keep both records rather than overwrite the earlier reading. Government food-safety pages are reorganised regularly and manufacturer documentation moves between support portals constantly, so each source also records what a reader must go and confirm and what to do if it turns out to say something narrower.

What we do with other people's content

We do not reproduce manufacturer manuals, cooking charts or recipes. We record structured facts and where they came from, in our own words, and point readers back to the primary source.

We do not scrape or republish copyrighted material. Where a source's licence restricts use, that restriction is recorded on the source and travels with every record citing it.

Where a manual is the best answer to a question, we say so and tell you how to find yours rather than hosting a copy of it.