Review intelligence
We read the reviews.You get the verdict.
Review Machine reads across professional reviews, buyer ratings and forum threads about one product, place or title, then reports what almost everyone agrees on, where reviewers split, and who should skip it. Every important observation keeps the source it came from.
- Sources
- Named and dated
- Review text
- Cited, never copied
- New verdict
- One a week
What a review keeps apart
Five kinds of claim, and never one number on top of them.
An aggregate score buries the difference between a measurement, a pattern, an opinion and a guess. A Master Review reports all five separately and labels which is which, so you can throw out the parts you do not trust and keep the rest.
Verified facts
What the thing actually is: the specification, the price, the running time, the room count, the release date — whatever the official source publishes. Dated, attributed, and never inferred from a reviewer's summary of it.
Recurring observations
What dozens of reviewers noticed separately, reported with how often it appeared and across how many sources. A complaint named three times in six hundred reviews is reported as three times, not as a pattern.
Individual opinions
The vivid single review is still one person's experience of one evening, one stay or one week. It is described for what it adds, counted as one, and never allowed to speak for the room it was taken from.
Editorial interpretation
Where we weigh one source above another, or argue that a complaint matters more than its frequency suggests. Marked as ours, so you can disagree with the judgement and still keep the evidence under it.
AI synthesis
Reading across hundreds of reviews is machine-assisted and we say so plainly. The synthesis is written in our own words, never pasted out of somebody's review, and it carries its citations with it.
How one is made
Collect, filter, extract, weigh, verdict.
Five passes in the same order every time, which is what makes two Master Reviews comparable with each other rather than two essays about different things.
001
Collect
Every review we can reach, plus the official source for the facts — a specification sheet, a menu, a room list, a running time. Then established publications, large reputable review platforms, forum threads for the qualitative signal, and video reviews where they can be analysed reliably. Source and date go down for each one.
002
Filter
Out go the duplicates, the reviews of a different model, a different branch or a different cut, and anything that reads as incentivised. What survives is the sample, and its size is reported rather than assumed.
003
Extract
The same fields out of every review: rating or sentiment, who the reviewer is and what they wanted from it, recurring praise, recurring complaints, the outliers, and anything that flatly contradicts the rest.
004
Weigh
Themes are counted and dated. Recent reviews carry more weight where the thing has changed — a firmware update, a refurbishment, a new chef, a second season — professional and public reviews stay in separate columns, and a thin or lopsided sample is reported as thin rather than rounded up into agreement.
005
Verdict
Written out in our own words: what the internet agrees on, what it argues about, who it suits and who should walk away. The sources sit at the end, so any line of it can be checked.
The weekly verdict
One verdict a week.
Every week, the most useful Master Review we finished: what the internet agrees on, the complaint that kept appearing, and who should skip it. Nothing else in the email. One click stops it, and the list is never sold or shared.
The reviews are already written. Somebody has to read them.
Master Reviews are free and there is nothing here to buy. Start with the archive, or have the week's most useful verdict sent to you and read the rest when you need it.