How we score
We collect reviews from professional publications, convert each verdict to a 0–100 score, and average them; we publish a score only when at least 4 reviews of the exact model exist, and label a score from just 4 reviews as an early score. With only 2 or 3 reviews we show a provisional average, clearly marked and never ranked.
Which reviews count?
- Professional reviewers who test products themselves: staffed publications such as Wirecutter, CNET, Tom's Guide, TechRadar and TechGearLab, plus a short, hand-checked list of established independent testers, such as Vacuum Wars, Project Farm and DC Rainmaker, including their YouTube reviews. Each independent reviewer is approved by hand: a named reviewer with a track record, their own testing, and real negative verdicts.
- No video counts if the maker of the product paid for it.
- No customer reviews, Reddit, unvetted YouTube channels, sponsored posts, manufacturer content or reviews from shops that sell the product. A shop has a reason to like what it sells, so we leave it out even when the review is good.
- Only the exact model. A review of the Branch Ergonomic Chair Pro does not count for the Branch Ergonomic Chair.
- One review per publication. Regional editions (for example PCMag UK and PCMag Australia) count as the same publication. We prefer a dedicated review over a buying guide, and the newest over older ones.
- Full reviews and tested buying guides only. We skip first looks, hands-on previews, "X vs Y" comparisons, news posts and index pages.
- Every exclusion is checked in code before a page is published, and reviews of a sibling model (say, the Kindle Scribe Colorsoft on the Kindle Colorsoft page) are removed by hand.
How is the score worked out?
When a publication gives a number (4 out of 5 stars, 8 out of 10), we convert it to 0–100. Many publications give no number. Then an AI model (Claude, made by Anthropic) reads the verdict and estimates one on the fixed scale below, and we mark that score "est." on the page. We then check those estimates against ratings the outlets actually print (see How do we keep our estimates honest?).
Where we use AI. AI does the reading at scale: it finds each review's score or verdict, estimates a score where none is printed, and writes the one-line summaries and the head-to-head notes in our own words. The rules for which reviews count, the calibration checks and the links to every source are there so you can verify any of it.
| Score | What the review says |
|---|---|
| 90 | Unreserved top recommendation. The highest we estimate, before the correction described below. |
| 85 | Clearly recommended. The most we give a product that only appears in a buying guide. |
| 80 | Recommended, with real caveats |
| 70–75 | Mixed: good, but the reviewer doubts the value |
| 60–65 | More problems than strengths |
| under 55 | Not recommended |
How do we keep our estimates honest?
An estimate is only useful if it's right, so we test ours. Whenever an outlet's own printed rating turns up for a review we had already estimated, we compare the two. Across 218 such reviews, our estimates came out slightly low:
| Our estimate | What the outlets printed (average) | Reviews compared | What we show |
|---|---|---|---|
| 70 | 71 | 12 | 71 |
| 75 | 78 | 10 | 77 |
| 80 | 83 | 103 | 83 |
| 85 | 90 | 78 | 89 |
| 90 | 95 | 15 | 92 (our ceiling) |
So we raise our estimates by the amount we measured, and never above 92: the very top of the scale is left to ratings an outlet actually printed. When we later find an outlet's printed rating, it replaces our estimate outright.
Do we change printed ratings?
Almost never. A printed rating is converted straight across: 4/5 is 80, 8/10 is 80, 92/100 is 92, and the original rating appears on the review. We checked that this is fair. We took outlets that rate coarsely (stars, or whole marks out of 10) and outlets that rate on a fine scale (out of 100, or with decimals), and lined up their ratings rank for rank on the same products:
| Coarse rating | Same rank on a fine scale | Critics who recommend it |
|---|---|---|
| 3.5/5 or 7/10 (70) | 70 | 3 of 26 |
| 4/5 or 8/10 (80) | 80 | 118 of 127 |
| 4.5/5 or 9/10 (90) | 90 | 67 of 67 |
| 5/5 or 10/10 (100) | about 96 | 33 of 33 |
The straight conversion holds everywhere except the very top. A perfect star rating covers a range of "excellent" rather than a single point, so a 5/5 or 10/10 counts as 95, and the review says so ("Rated 5/5 · counted as 95"). That is the only change we make to a printed rating. Based on 259 coarse and 581 fine printed ratings, checked 2 Nov 2026.
Every product page shows how many of its scores are printed and how many estimated, and the average of the printed scores alone, so you can always compare. Video reviews are estimated by a separate process and corrected with their own curve, fitted the same way.
The critic score is the plain average of the reviews, rounded to the nearest whole number (halves round up, so 86.5 becomes 87). "Recommend" counts the reviews whose own verdict is a recommendation, not the reviews above some score.
What do "Critics agree" and the other labels mean?
Next to every score we say in plain words how the critics' scores sit together. The labels are set by fixed rules from the scores alone, and they never change the score itself:
- Critics agree: every score is within 12 points of the others.
- Broad agreement: a wider spread, but no camp of dissenters.
- One critic disagrees (or is far keener): one review sits 18 or more points away from the others, while the rest agree with each other. We name that critic and show what the others average, so you can see why the score is where it is. Their review still counts in full.
- Critics split: scores 30 or more points apart with no single outlier, so the critics fall into camps. Read the individual reviews before buying.
Common questions
Do you include YouTube reviews?
Yes. Hand-vetted channels that test products themselves count, such as Just Josh, DC Rainmaker, Vacuum Wars, Jarrod'sTech. Sponsored videos never count, and a reviewer with both a site and a channel counts once per product.
Why isn't Consumer Reports included?
Its terms forbid commercial use of its ratings. RTINGS and CHOICE are left out for similar reasons.
What about reviews that give no score?
An AI model estimates one from the verdict, and the page marks it “est.” We test those estimates against ratings the outlets print and correct them.
Do you include customer reviews?
No. No customer reviews, Reddit, forums, shops or manufacturers.
Does one reviewer ever count twice?
No. Each reviewer counts once per product: a publication's regional editions and its YouTube channel are one reviewer, and a writer who reviews the same product for two outlets counts once.
How many products and reviews do you cover?
999 products, 5,269 professional reviews from 213 outlets and channels.
How do we make money?
When you buy through a store link, we may earn a commission at no extra cost to you. It never affects scores: scores come only from the reviews. As an Amazon Associate we earn from qualifying purchases.
Found a mistake?
Tell us at [email protected] and we will fix it. Every correction is listed on the corrections page.