How we research
Reviews are researched with AI and checked by a named editor. Every score traces back to cited evidence.
What goes into every BuyScore
Every review starts from public evidence: independent expert reviews and measurement sites, manufacturer documentation, official safety and security records, dated price data and public owner discussions. We read those sources, summarise them in our own words and link to them.
How AI is used
AI researches each product from public sources and drafts the research file, the criterion scores and the verdict. A second AI pass checks every citation. A named editor then opens the sources behind every score, the verdict, the safety check and every price, and signs off the page. Every review says who checked it and when.
The AI never calculates the BuyScore. It returns a score for each criterion with a reason and sources; the site does the maths from the category weights.
Evidence Confidence
BuyScore says how good a product is for the job. Evidence Confidence says how sure we are. It is a points total from five checks: expert coverage, official data, owner discussions, agreement between sources and how recent the evidence is. Fewer than two independent sources always means Insufficient, and then there is no BuyScore.
What "hands-on" means
We only mark a product as hands-on when we used it ourselves, for example a free software trial or a product we bought. We never claim testing we did not do. Most of our reviews are research reviews built from the sources cited on the page.
What we never do
We don't copy other sites' scores or star ratings, publish percentages of owner reviews without a proper dataset, or call a product "safe" or "secure". Commission rates are never an input to a score.
Evidence Confidence levels
| Points | Level | Dots | Effect |
|---|---|---|---|
| 9–10 | Very high |
|
Eligible for every pick label |
| 7–8 | High |
|
Eligible for every pick label |
| 5–6 | Moderate |
|
Picks allowed except "Best overall" |
| 3–4 | Limited |
Limited evidence
|
Score shown, no picks |
| 0–2 | Insufficient |
|
No score |
Points come from five checks: expert coverage (0–3), official data (0–2), owner discussions (0–1), agreement between sources (0–2) and recency (0–2), plus one point when we used the product ourselves. Fewer than two independent sources is always Insufficient.
Sources we use, and how
| Source | What it's for | What we never do |
|---|---|---|
| Expert reviews and measurement sites | Summarised in our own words, cited and linked; counted for Expert Consensus | Copy their scores, charts, photos or logos |
| Security test labs and audits | Security criteria, with test dates | Present an old test as current, or call a product "secure" |
| Public owner discussions | Common themes in our own words, with links | Publish percentages or counts, or use their star ratings as ours |
| Manufacturer and vendor documentation | Claimed specs, plans, certifications, SLA terms | Treat claims as verified without other evidence |
| Safety and vulnerability records | "No recalls found in [database] as of [date]" | Write "Recall: clear" or "Safe" |
| Dated price data | Value, value guidance and Cost Reality | Hand-typed prices that go stale |