How we rate products
CoolCompare uses published evidence to make product differences easier to understand. We do not convert every manufacturer claim into one supposedly scientific score, and we never present an editorial calculation as a customer review or a laboratory test. A score can organise a shortlist; the underlying value, source, test context and room requirement remain the decision evidence.
Last reviewed: 28 August 2026
Four confidence labels
Each metric receives a comparability label before it can influence a summary. Directly comparable means the unit and definition are compatible, such as physical dimensions or rated input recorded on the same basis. Broadly comparable means the value can orient a decision but needs a caveat. Manufacturer-specific test means we preserve the maker's stated method and do not use it to declare a cross-brand winner. Unknown means no value is inferred and no points are awarded.
- Directly comparable: compatible definition and unit.
- Broadly comparable: useful direction with a visible limitation.
- Manufacturer-specific test: source claim retained with its method.
- Unknown: excluded rather than estimated.
What can enter a specification score
Cooling capacity, rated power, physical size, weight and clearly defined functions may contribute when source evidence is present. Sound, efficiency and room guidance require additional care because brands can publish different modes, distances, test conditions or room assumptions. A missing value is never treated as zero or quietly replaced with a category average.
Weights are chosen for the stated use case. Bedroom-oriented ordering can give more weight to compatible maximum sound data; a running-cost view can order verified rated input. Capacity is not automatically rewarded because an oversized product is not intrinsically better for a small room. Every commercial shortlist explains its ordering rule on the page.
What the score does not mean
A CoolCompare score is not proof that we physically tested the appliance. It is not a reliability forecast, an owner-satisfaction rating or a guarantee of comfort. Manufacturer specifications cannot describe every factor that matters in a real room, including compressor tone, hose sealing, solar gain, maintenance, controls and local support.
Where real-world user submissions become available, they remain a separate, moderated dataset. We show aggregate claims only after a defined minimum sample and do not mix them invisibly into manufacturer data. A recall or material safety notice overrides normal commercial presentation rather than merely subtracting a few points.
Commercial independence and corrections
Retailer commission cannot create a rating, purchase a position or hide a stronger alternative. Sponsored placement must be labelled and kept separate from the editorial order. A retailer link can disappear when its evidence becomes stale without changing the stored product specification record.
Readers, manufacturers and retailers may submit corrections with supporting evidence. We verify the exact model and source, update the reviewed date when the change is material and preserve the distinction between corrected source data and our calculated outputs. Disagreement alone does not trigger a change; traceable evidence does.