Methodology

What this site is

Better Selects is a small US site that watches Amazon US prices and writes up the products worth buying at them. It currently carries 36 guides covering 282 products. It is funded by Amazon affiliate commissions, which is set out in full on the affiliate disclosure page.

We have no lab and no test bench. We have not opened these products, measured them, or lived with them. What we do have is the listing data, a set of rules about which parts of it can be repeated, and a checker that reads our own sentences back against the source. This page describes all three, including where they stop.

Where the numbers come from

Every price, star rating and rating count on this site was read off a live Amazon US listing. Candidates are gathered from Amazon search, filtered, and then re-checked on the individual product page. The product page wins on every field, because search results roll variants together — a search listing will report the rating count of a whole family of colours and sizes, and sometimes quotes a per-unit price for a multipack. Where a product-page check is blocked, the earlier figure stands with its earlier date rather than being blanked or quietly refreshed.

Products with very little feedback do not make the shortlist at all. The floor is roughly a thousand ratings, because below that there is no failure pattern to report, only a small number of early buyers.

The rule that throws out fake discounts

A struck-through “list” price on Amazon is typed in by the seller. Nobody audits it. The easiest way to run a site like this one is to repeat that number, subtract the asking price, and print the difference as a saving.

We discard it. If the struck-through price is more than three times what the product is actually asking, we treat it as the seller’s own invention: the site will not print it, and it cannot become a percentage anywhere — not in a badge, not on the homepage table, not on a card. A cut under 15% is not called a deal either, because it is inside the range a price wanders through on its own. Both thresholds live in exactly one function, so no new part of the site can route around them.

Where a product is genuinely only cheap against an inflated list, the entry says so rather than leaving it out.

Every price carries the date we saw it

Nothing here says a product costs $49 without saying when we saw $49. Amazon moves prices hourly, and any other format would be a claim we cannot keep. The writing follows the same rule: the guides do not say “currently” or “right now”, because the page is not live at the moment you read it — it says what the price was on the date printed beside it.

When a pick sells out, it keeps its entry and loses its price. The product stays on the page, because it is still the right answer and it will come back, but we stop making a claim about what it costs, and it drops to the bottom of the list.

Words and numbers are kept apart

The prose and the data are written separately and then merged. The editorial file supplies sentences and nothing else; the harvest supplies every figure. A pick that appears in the writing but not in the harvest is dropped instead of published. There is no path by which a price, a star rating or a rating count on this site could have come from a writer’s memory.

What the writing is allowed to assert is deliberately narrow: the product’s real numbers, and what its own title states — capacity, wattage, size, model number, count. General category knowledge is fair game, because it is true independently of any one unit: how noise cancelling works, why every US space heater is 1500 watts, what a coil-spring seat means. What is never allowed is a specific claim about a specific unit that we cannot support. We do not quote reviews we have not read, name a percentage of owners, cite a failure rate, describe a teardown, or say we tested anything. Where a product is too new to have a known failure pattern, the entry says that, rather than guessing at one.

The claims checker

Rules that live only in a style guide get broken. So the rating and review-count claims in our own prose are checked by a script against the harvested facts before anything goes out, sentence by sentence, across every guide on the site.

It looks for seven things:

Explicit counts — “28,606 ratings”, “nearly 14,000”, “17,000-plus”, “a six-figure rating pool” — including whether a hedge like nearly, over or roughly is doing honest work for the real number.
Star ratings, separated from version numbers and measurements that look like them.
Superlatives: the most-reviewed, the highest-rated, the thinnest record, mid-pack. The claim has to be held by the product it is written about, and a tie has to be described as a tie.
Margins — “by a wide margin” is checked against the runner-up, not just the leader.
Short-history notes. “Too new to have a failure pattern” has to sit on a product with fewer than 3,000 ratings, and the thinnest record on a page that uses such notes has to carry one.
Comparisons: “a tenth of the other one’s”, “twice as many”, “second only to”, and a quoted figure attributed to a named product.
Claims that two listings share a rating count, which is usually a variant mix-up.

Anything that fails is an error, not a warning, and it is fixed before the guide is published. The checker is conservative by design: where a claim is scoped to a subset it cannot identify — “the longest record here for a straight dryer” — it declines to judge rather than passing it, and that sentence is read by a person.

What we cannot tell you

This is the part most sites leave out, so it is worth being blunt about.

We do not test products. There is no lab, no meter and no sample. Nothing on this site is a measurement we took.
The ratings are Amazon’s, not ours. We repeat a star average and a rating count accurately and date them. We cannot tell you how many of those ratings are genuine, and neither can anyone else outside Amazon.
Our price history is days old, not years. We know what a product asked when we checked it and what it asked the previous time. We do not hold a multi-year price curve, so we cannot tell you whether today is the lowest this item has ever been — only what we have seen it at.
A listing can change underneath us. Amazon lets a seller alter a bundle, swap the fulfilling seller, or replace the model behind an existing product page. A link that was right when we checked it can point at something slightly different later.
Amazon US only. Prices, availability and even which model is sold differ by country. Nothing here describes another storefront.
Pages refresh on a schedule. Prices are re-read periodically rather than continuously, so the date beside a price is the honest reading of it, and the price on Amazon at checkout is the only one that is real.

Corrections

If a number here is wrong, it is a defect and we want to know. Write to hello@betterselects.com with the page and the figure, and it gets fixed. Nothing on this site is worth keeping if the numbers are not right.