Sources we use, in order of trust
Every profile names the exact US model and links to its sources. Product facts are manufacturer statements unless labelled otherwise. We rely on these sources, in this order: (1) the exact US model’s manual or owner’s guide and the manufacturer’s US product page or spec sheet; (2) manufacturer support, warranty and parts pages and dated manufacturer announcements; (3) government sources, such as CPSC recall records for recall matching and FDA, USDA and NCI guidance in food-safety articles; (4) our own calculations, which are labelled with their formula and assumptions.
When the manual labels something the product page leaves ambiguous, such as which dimension is the depth, the labelled manual figure settles it. When official sources disagree and nothing resolves the conflict, we withhold the value or show it as unconfirmed and say why, rather than picking one. Figures from another country’s version of a model are not used for the US model. We do not use third-party reviews, retailer star ratings or other sites’ test results as product facts.
How manuals are read and checked
In our 2026-10-15 manual pass we checked all 48 manuals then linked from profiles and read 46. The two Gourmia manuals were blocked by an automated bot check on the manufacturer’s site and were not read. Readers extracted clearances, materials, dimensions and other facts from each PDF without seeing the values already in our records, and we keep the manual page reference for each fact found.
We then checked a seeded random 20% sample of the positive claims extracted (15 of 77, covering clearances, materials and dimensions) blind: a fresh reader received only the manual and the claim. All 15 were confirmed. Statements recorded as “not stated in the manual” were not part of that check. A blind check of a sample reduces extraction errors; it does not prove every record is correct, so corrections are welcome.
How lists and the shortlist are ordered
Use-case pages (for example, air fryers for one person) list every documented model that meets the stated criterion, such as a single basket of up to 5 advertised quarts. They are not rankings: models appear in our catalog order, which is largely grouped by brand, and position says nothing about quality. A fact we do not have never qualifies a model. Models with a matched recall or an unresolved purchase status are left out of these lists.
The homepage shortlist quiz does rank. It first removes models that do not fit your answers: household size sets an advertised-capacity range, the basket layout must match, and oven-style models, models without a stated quart capacity, and recalled, unresolved or discontinued models are excluded. Each remaining model then gets an additive score. Your stated priority (smallest documented footprint, most advertised capacity, or documented dishwasher-safe parts) has weight 3. Closeness to our guide capacity for your household (about 3, 5, 7 or 9 quarts, plus 2 for two-basket models) has weight 1.5. The extra you choose, more documented cooking functions or official support and parts links, has weight 1. Price has weight 2 when you choose a budget, but it only counts verified current prices; none are connected today, so price does not change the order.
A fact we do not know scores zero for the part of the score that needs it: an unknown footprint is never treated as compact. Ties go to models with a linked manual and documented footprint. The three results include at most one model per brand while another brand still matches. Affiliate compensation is not an input to any list or score.
How often pages are updated
There is no fixed refresh schedule. We recheck records in research batches, such as the manual pass described above, and when a reader sends a correction or a manufacturer publishes a new document. The “Specifications checked” date on a profile is when its main source was checked. Facts added in a later batch carry their own dates in our research records, so some details may be newer than that date. Prices are not shown until a licensed price feed is connected, and recall coverage is a selection of records, not every recall.
Who does the work, and where AI is used
AI agents do much of the research legwork. They find official documents, read manuals and product pages, extract facts into structured records and draft page text. The site owner reviews research batches and copy before publication. Automated checks run before release: they validate the catalog, check links, and fail a build that shows placeholder values or internal research notes on a page.
Nobody has cooked with these appliances for this site. Profiles are researched assessments, not hands-on reviews, and we do not publish our own star ratings.
What we do not know
Unknown specifications remain unknown. We do not infer usable cooking area from advertised quarts, safe operation from external dimensions, or release date from the day we found a listing.
Tools use published facts or clearly labelled inputs that you can edit. Electricity calculated from rated power assumes full power throughout the runtime. Measured energy is preferable. To report an error, see sources and corrections.
Implementation described 2026-10-15. No hands-on testing claimed.