How We Test Products
An honest account of our research process, including what it does not cover.
Our process
-
Define the category and the buyer
Before looking at products we write down who the article is for and what they are trying to achieve. A portable air conditioner for a rented flat and one for a workshop are different problems, and a ranking that ignores that is useless to both readers.
-
Build the long list
We assemble every product that plausibly serves that buyer, typically between fifteen and twenty-five models. This includes products with no affiliate programme, and products from brands we have criticised before. The number of models considered is published at the top of every article.
-
Establish the criteria that decide the category
Every category has a small number of factors that actually determine whether a product is good, and a large number of specifications that manufacturers emphasise instead. We identify the former first. In air purifiers it is clean air delivery rate and five-year filter cost; in car seats it is how hard the seat is to install correctly. We publish these criteria in the buying guide section of each article.
-
Research each shortlisted product
We read manufacturer technical documentation and manuals rather than marketing pages, check the applicable safety and performance standards, review published independent and laboratory testing where it exists, examine regulatory notices and recall databases, and read large volumes of long-term owner reports to identify failure patterns that only appear after months of use.
-
Hands-on assessment where we have it
Some products in our categories we have used ourselves over extended periods. Where our assessment of a product includes hands-on use, we say so explicitly in that product's write-up. Where it does not, our conclusions rest on the documentary research described above, and we do not describe results we have not observed.
-
Score and rank
Each product receives an editorial score out of ten, weighted by the category criteria established in step three. Scores are comparable within an article and not across articles: an 8.5 in blenders and an 8.5 in car seats were reached against different criteria.
-
Review and publish
Every article is checked by a second person against the sources before publication. Safety-critical categories, particularly anything for infants, are checked against the current guidance of the relevant standards or paediatric body at the time of publication.
-
Revisit
We re-check every article on a scheduled cycle and immediately if a product is recalled, discontinued or substantially revised. The last review date appears at the top of every article, and it is a real date rather than a script that prints today.
What our scores mean
- 9.0 and above. Excels against the criteria that matter in the category. We would recommend it without qualification to the reader the article is written for.
- 8.0 to 8.9. Very good, usually with one clear trade-off that we name explicitly.
- 7.0 to 7.9. Competent, and the right answer for a specific reader rather than a general one.
- Below 7.0. We generally do not publish it. If a product appears below this and we have included it anyway, the article explains why.
What our process does not cover
Stating limits is part of being useful.
- We do not operate a certified laboratory and we do not publish instrumented measurements as though we produced them. Where we cite test data, it comes from a named independent source.
- We cannot test crash performance, electrical safety or fire resistance. For those we rely on the mandatory standards each product must meet and on published regulatory testing.
- We cannot assess long-term reliability of a product released this month. Where a product is too new for failure patterns to have emerged, we say so rather than guessing.
- Prices and availability change faster than we can publish. We do not quote prices in articles for that reason, and we send you to the retailer to see the current one.
Conflicts of interest
We buy the products we assess or use publicly available documentation. We do not accept free samples in exchange for coverage, and where a manufacturer has supplied a product for any reason we disclose it in that article. No brand sees an article before publication, and no brand can change a score.
Methodology last reviewed: August 1, 2026