Methodology

Appliance Compare publishes data about UK home appliances drawn from four sources: the EU EPREL energy-label register, retailer affiliate feeds, community sentiment harvested from public review platforms, and editorial review by our team. Each is described below with its sources, refresh cadence, what it is used for on the site, and its known limitations.

Specifications and energy data

Source. The EU EPREL register, the regulatory database manufacturers must populate to sell an energy-labelled appliance in the European market. We ingest the relevant category endpoints (currently refrigerating appliances; televisions, washing machines and dishwashers are being added) and join the resulting records to our canonical SKU table by manufacturer model number.

What we publish from it. Width, height, depth in millimetres; total and (where applicable) compartment capacities; energy class on the current 2019 A-to-G scale; annual energy consumption in kWh; noise level in decibels and the corresponding noise class; freezer star rating; climate class; and a small set of feature flags the regulation requires.

Refresh. Monthly bulk pull of the relevant categories; an additional pull when an EPREL record version increment indicates a change. Every API response and product page surfaces the last EPREL sync time so you can see how fresh the spec data is.

Limitation. EPREL covers products that are EU-registered. Some UK-market or US-market SKUs may not appear. Where a manufacturer model number cannot be unambiguously linked to an EPREL record at high confidence, we fall back on a family-sibling match (the same model line, a slightly different colour or trim) and flag this in our internal audit table. Inheritable specs such as energy class are generally identical across colour variants; specs that legitimately vary are not propagated.

Prices and retailer availability

Source. Affiliate product feeds from major UK retailers, delivered through the Awin publisher network. Each feed contains the retailer's current product listings, prices, stock state and a deep-link URL.

Refresh. Daily import overnight, with the materialised view of "current price" refreshed at the end of each import. The "Prices updated" timestamp in the site footer reflects the latest successful import.

What we publish from it. The lowest in-stock price across matched retailers, the cheapest retailer's name, the count of distinct retailers carrying the SKU, and a "View" link that routes through the affiliate network to the retailer's product page.

How "cheapest" is decided. Among the listings we have for a given canonical SKU across retailers, we prefer the listing that is currently in stock; within in-stock listings we pick the lowest amount. A historically lower out-of-stock price is not surfaced as the headline figure because you cannot buy it.

Limitation. Retailer coverage is being expanded over the summer of 2026. A product without a price has not been priced because no participating retailer has a current listing matched to it in our catalogue — not because we have hidden a price.

Community sentiment (Phase 1 of the reliability moat)

Source. Public posts, comments and reviews from UK-relevant consumer discussion platforms (currently YouTube comments on appliance review videos, supplemented over time by other public sources). Posts are ingested via official platform APIs where available; we do not scrape behind login walls or behind robots.txt blocks.

What we do with it. Each post is processed by an AI extraction step (Anthropic's Claude Haiku model with a strict tool-call schema) that identifies the brand, model and category being discussed; the writer's sentiment toward that product on a five-point scale; and any specific failure modes mentioned, drawn from a controlled vocabulary of twenty-one canonical labels (compressor, motor, drum bearing, control board, door seal, water leak, noise, software bug and so on).

How it is aggregated. Signals are rolled up at the brand level (currently the primary scope), and at the per-model level where we have enough signal density for a particular SKU. Brand-level scores require at least ten distinct signals; per-SKU scores require at least three. Aggregations below those thresholds are flagged "preview / unreviewed" or hidden.

What we call it. "Community Sentiment", explicitly — not "Reliability Score". Self-reported online sentiment has well-known selection bias: people post when an appliance breaks more readily than when it quietly works. Calling this a reliability score would overstate the strength of the evidence and is not something we are willing to do. Where we display sentiment data, we display the sample size and the time window it covers so you can judge confidence yourself.

Limitation. Most comments mention a brand without naming a specific model. The signal is therefore richest at the brand level and sparse at the per-SKU level. Per-SKU sentiment will improve as our SKU coverage grows and as the structured reliability survey (Phase 2) lands.

Reliability survey (Phase 2 of the reliability moat)

Beginning in autumn 2026 we will run a recurring UK appliance reliability survey: a structured questionnaire put to real owners about ownership length, failure modes experienced, repair experience and overall recommendation. The first report is planned for late 2026 / early 2027 and will be published with methodology, sample sizes, confidence intervals and brand-level rankings — the same statistical-disclosure standard used by the established consumer reliability publications.

Until the survey has been running long enough to produce statistically meaningful results, the site will continue to use the community-sentiment score described above as the reliability surface, clearly labelled as such.

Total cost of ownership

What we publish. A 10-year TCO figure on every product with a known price, computed as: cheapest current price + (annual kWh × 10 × current UK electricity tariff).

The annual kWh figure comes directly from EPREL for fridges, whose label declares a continuous-use annual consumption. For washing machines, dishwashers and washer-dryers, the label declares per-cycle consumption; we multiply by a CSE-aligned assumed cycles-per-year (220, 150 and 280 respectively). For televisions, the label declares power draw in watts; we multiply by an assumed 4 hours/day of operation, in line with the EU 2019/2013 labelling methodology. These cycles-per-year and hours-per-day assumptions are stored in a single table and revised in one place when CSE or industry guidance updates.

The electricity tariff is the current Ofgem-published default-tariff unit rate for the country (Great Britain). It is read from our tariffs table at request time, so the figure reflects today's tariff rather than a year-old assumption.

Editorial verdicts

A subset of products carry an editorial verdict written by our team. The verdict represents our judgement on the product's strengths and weaknesses compared with peers of the same category, capacity bracket and price bracket. Verdicts are written before any commercial data is consulted and are not changed in response to commission rates or retailer requests.

Where we have written a verdict we display it with our editorial score; where we have not, we say so rather than infer one from the spec data alone. As of June 2026, editorial coverage is a small fraction of the catalogue and is being expanded.

Rankings and default sort

The order in which products appear on a category page is determined by an intelligent default sort the matching engine selects based on the filters you have applied:

• If you have set a price ceiling, results are sorted by editorial score per pound spent — the best value at your budget.
• If you have applied a quality or feature filter (energy class minimum, editorial tier, etc.) without a price ceiling, results are sorted by editorial score.
• If you have not applied either, results are sorted by editorial pick for the size class you are looking at.

You can override the default at any time using the sort controls on the results page. The default-sort decision is also surfaced in the page header so you can see which rule the engine applied.

What we do not do

• No paid placements. Brands and retailers cannot buy a ranking, a slot, a flag or a badge.
• No sponsored rankings. Awin commission rates are not an input to the sort order.
• No relabelling. Community sentiment is community sentiment, the reliability survey is the reliability survey, and editorial is editorial. We do not blend the three under a single label.
• No editorial verdicts written to a commercial brief.

Citing this data

If you are writing about UK appliance specifications, prices or community sentiment and would like to cite a number from this site, please reference "Appliance Compare, <date>" and a link to the specific product page. For aggregated cuts — "the highest community-sentiment fridge-freezer brands in the UK as of June 2026", etc. — email hello@appliancecompare.uk and we will give you a defensible quote with the sample size, window and methodology version attached.