Most write-ups about consumer reviews start with a number and stop there. So before we draw any conclusions, here is the honest shape of what we are actually working with as of June 6, 2026: 549,363 reviews, every one of them scored for sentiment, roughly nine in ten carrying real written text rather than a bare star rating, spread across more than 21,000 physical locations, with the oldest entry dating back to 2008 and the newest landing yesterday.
That is a big enough corpus to say something real about how people actually talk about the brands they visit. But the interesting part is not the size. It is the shape — and the shape has three lessons that change how you should read any "average rating" you have ever seen.
1. Sentiment leans positive — but a third of it is negative
When you score every review on a scale from −1 (strongly negative) to +1 (strongly positive), the corpus averages out to a mild +0.13. People are, on balance, slightly happier than they are unhappy. But the average buries the split that matters:
Fifty-five percent positive sounds comfortable until you sit with the other number: one in three reviews is actively negative. For a brand operator, that is not background noise — it is a standing third of your published feedback telling a story you would rather not be told. The useful signal is rarely the headline average. It is whether that 33% is concentrated in particular locations, particular months, or particular complaint themes.
2. Ratings are barbell-shaped, and the average describes almost nobody
Here is the finding that surprises people most. If you assume star ratings cluster around a happy middle, the data disagrees emphatically. Reviewers overwhelmingly pick an extreme:
| Star rating | Share of reviews |
|---|---|
| 5 stars | 44% |
| 4 stars | 11% |
| 3 stars | 8% |
| 2 stars | 7% |
| 1 star | 30% |
Forty-four percent of reviews are five stars. Thirty percent are one star. That means nearly three out of every four reviews sit at the very top or the very bottom of the scale, with the 2-, 3-, and 4-star middle accounting for barely a quarter between them. The headline average across the whole corpus is 3.33 stars — a number that almost no individual reviewer actually chose.
People do not write reviews to record a mild, balanced opinion. They write when something delighted them or when something went wrong. The middle of the scale is mostly empty, which is exactly why a single average is the least informative way to read review data.
The practical takeaway: if you are tracking a brand and its average rating ticks from 3.3 to 3.4, that does not mean "everyone got a little happier." It almost always means the mix between the 5-star camp and the 1-star camp shifted. Watching the poles — and what drives each one — tells you far more than watching the mean.
3. Stars and text do not always agree
If ratings were a perfect proxy for sentiment, the two distributions would line up. They do not. The text-level sentiment reads 55% positive, while only 44% of reviews are five stars. The gap lives in the language people use: a four-star review can carry a glowing paragraph, and a five-star review can quietly mention a problem in passing. A blunt one-to-five scale flattens all of that nuance.
This is the whole case for reading the text, not just the score. Natural-language scoring picks up the “great food but the app crashed twice” reviews that a star rating rounds away. Across half a million reviews, those rounding errors are not edge cases — they are a meaningful slice of the signal.
The lesson hiding underneath: a single snapshot lies
There is a subtler trap in this kind of data, and it is worth being honest about because most "consumer sentiment" charts ignore it. When we look at month-by-month averages, the recent numbers move around — not always because consumers changed their minds, but because which locations got refreshed that month changed. Coverage shifts. Sources retire. A month with fewer, newer reviews can read artificially rosy.
That is not a flaw to hide; it is the central reason review data has to be handled carefully. Two things make it trustworthy instead of misleading:
- Provenance on every row. Each review in the dataset records where it came from, which scoring method produced its sentiment, and when it was captured. Three different methods score the corpus today, and every review is tagged with the one that scored it — so a number can always be traced back to how it was made.
- Point-in-time capture. A one-off scrape gives you a single, un-auditable photograph. Capturing the same locations repeatedly and stamping each observation with a timestamp gives you a series — and a series is the only thing that lets you tell a real shift from a sampling artifact.
Put simply: the value is not in owning half a million rows. It is in being able to defend every one of them — what it says, where it came from, and when it was true.
How we think about it at ReviewSignal
ReviewSignal is a consumer review intelligence platform for monitored-chain workspaces. We track a curated universe of consumer brands location by location, score the sentiment in their reviews, and surface where a brand is moving, where the move is concentrated, and which themes are driving it — then let you export a clean snapshot you can put in front of a colleague.
The aim is not a magic number. It is an honest, traceable read on consumer feedback at scale: the distribution, not just the average; the text, not just the stars; the series, not just today's snapshot.
Want to see the shape of the data for the chains you care about? Tell us the brands you watch and we will show you the distribution, the drivers, and a snapshot you can share. Reach the team at team@reviewsignal.ai or get in touch here.