How we research, and how we pick deals

The short version

Refer a Deal is a research desk, not a test lab. For every buying guide we read the manufacturer's own documentation, normalise the specifications so they can actually be compared, quantify what verified owners report after months of use, and check the price against its own history. Each article carries a label telling you which method produced it. We never write as if we used a product we have not used.

What we actually do

1. We read the primary sources first

Manufacturer specification sheets, the full product manual rather than the marketing page, warranty terms, and support documentation. We also read the Q&A section of a product's listing, where sellers answer real buyer questions — that is often where a limitation shows up that the spec table leaves out.

2. We normalise the specifications

Suction quoted in Pa, in air watts, or as a multiple of some unnamed baseline. Runtime measured at different power settings. Noise measured at different distances. Numbers that look comparable usually are not. We convert what can be converted into a single set of units, and we mark anything that cannot be converted as unverified rather than printing it side by side as if it were equivalent.

3. We quantify what owners report

This is the part we spend the most time on. We work through the verified-purchase reviews on a product's listing and tag each one by subject rather than by star rating — navigation faults, battery behaviour, consumable costs, support experience, and so on. That produces claims we can actually stand behind, such as “[N]% of the reviews we read mention [issue]”, and it includes the numbers that count against a product. Where we work from a sample rather than the full set of reviews, we say so and give the sample size.

4. We track the price, not the discount claim

A “40% off” badge means nothing on its own. We compare the current price against the product's own 90-day range so you can see whether it is near its real floor or hovering above its own average. Then we tell you whether to buy now or wait.

What we don't claim

• We don't say we tested something we didn't. No article on this site will describe a product being used unless we actually had it in hand.
• We can't tell you how it feels. Hand feel, build impressions and long-term durability beyond what owner reports support are outside what research can honestly deliver.
• We don't have lab measurements. Where a number only exists because someone measured it on a bench, we credit whoever measured it and link to them.
• We don't fill gaps with confident prose. When the evidence is thin or contradictory, the article says so.
• Our images are illustrations, not photographs of the products. They are editorial artwork made for this site, and they are labelled as illustrations rather than presented as pictures of the thing you would receive.

Where our numbers come from

Every figure in a Refer a Deal article traces back to one of these sources. If you want to check one, this is where to start.

• Manufacturer documentation — specification sheets, manuals, warranty terms and support pages, for anything we state as a hard specification.
• Verified-purchase owner reviews — for reported problems, recurring complaints, consumable costs and what actually breaks.
• Price history — for current price against the 90-day range, and for judging whether a discount is real.
• Marketplace and category data — for how a listing actually behaves over time, which models are genuinely widely owned, and which questions buyers ask most.
• Independent professional reviews and test labs — credited by name, used to cross-check a claim rather than to replace our own sourcing.
• Owner video and forum reports — for durability signals that only show up months in. Attributed to the person who recorded them.

Data & evidence: what we counted, and how

Every number on this site traces back to a named dataset. This is the full list of what we counted, when we collected it, and what we left out — so a claim can be checked against the method instead of taken on trust.

Review-coding studies

Our data studies tag each verified-purchase review by subject rather than by star rating. The coded sample is smaller than the full review pool because we only code reviews whose text we can actually read.

Single-product owner samples

Price history

Deal verdicts and price charts compare the current price against the product’s own 90-day range. The window runs from the first to the last observed date inclusive, which is 91 calendar dates. We record the lowest, average and highest listing price observed in that window.

• Source — third-party price-tracking data (Keepa and similar), cross-checked against the listing.
• Range — 90 days ending on the “checked” date shown on each deal card.
• Excluded — list prices and “was” prices the product never actually sold at, and out-of-stock snapshots.

Ratings totals

Where an article cites a total rating count — for example “319k ratings” or “112k ratings” — the figure is the aggregate rating count shown on the Amazon listing at collection time. It is a snapshot, not a live count.

What these numbers do not mean

A coded sample is not the full set of reviews, and none of these figures is a product failure rate. They describe complaints within the reviews we actually read. Where a finding rests on a sample, the article states the sample size and the review pool.

What the labels on our articles mean

Every article here is a research-desk analysis — that is the baseline and is no longer labelled. The labels below tell you which additional checks were applied to the piece you are reading, so you can weigh it accordingly.

• Spec-verified — every specification quoted was checked against the manufacturer's own documentation.
• Price-tracked — the article compares the current price against the product's 90-day history.
• Hands-on tested — we had this product in hand. Used only when that is true, and the article will say how we got it and how long we had it.

How we decide a deal is real

Retailers run “sales” constantly by raising a price for a few weeks and then cutting it back to where it started. To filter that out, offers labelled “Near 90-day low” or “Below average” are checked against their own 90-day price history before they appear. Offers labelled “On sale” carry a current promotion (such as a coupon or an Amazon deal flag) but are not claimed to be at a 90-day low.

• The current price must be genuinely lower than the recent median, not just lower than a recent peak
• We ignore list prices that the product never actually sold at
• Stock that has been permanently discounted for months is not a deal — we archive it to the deal history section instead of featuring it

How we make money

Refer a Deal is funded by affiliate commission. When you buy through a link on this site, the retailer pays us a small percentage of the sale. This costs you nothing extra — the price you pay is the same either way.

What commission does not buy: a better score, a higher ranking, or a place on the deals page. We do not accept payment for a placement, we do not let a brand see an article before it is published, and a discount on a product we would not recommend is still a product we would not recommend.

Who writes here

One person writes the guides on this site and does the research behind them: Nora Whitfield. The background that makes this kind of research possible is on that page.

Corrections

If a price has changed, a specification is wrong, or a recommendation has aged badly, we want to know. Corrections are made in the article itself, and the date at the top is updated. Tell us about an error.