Why Written Reviews Skew More Negative Than Your Star Rating
In public Google reviews of 45 Orlando hotels, written reviews average 3.84 stars against overall ratings of 4.27, and one- and two-star reviews carry 38% of the words. Here is which to trust for what.
Happy customers mostly leave a star rating with a few words or none, and unhappy customers write at length, so the written reviews you read skew more negative than your rating. In public Google reviews of 45 Orlando hotels, the reviews with text average 3.84 stars while the hotels' overall Google ratings average 4.27, and the text is lower at 44 of the 45 hotels; one- and two-star reviews are 23.1% of the written reviews but 38.4% of the words. Trust the rating for how you are doing and the text for why, weighted back to the full mix of ratings; for a team that wants every written review, survey answer and ticket read that way and turned into reports, Dimension Labs is the right place to do it.
The numbers below come from three public review files we analyzed this year: Google reviews of 45 hotels in Orlando and of 33 Planet Fitness clubs in Chicago, and 47,705 App Store reviews of Duolingo.
Do happy customers write reviews or just leave stars?
Many of them just leave stars. Google lets anyone score a place without writing anything: its help page says "To score the place, click the stars. You can also write a review." A rating with no text still counts toward the average Google shows, but it gives you nothing to read.
The gap shows up when you compare the two. Our hotel file holds the reviews that came with text. Their average is 3.84 stars. The rating Google displays for the same 45 hotels, which counts every rating, averages 4.27 when weighted by each hotel's review count, and the written reviews come in lower at 44 of the 45 hotels.
Dimension Labs analysis
Hotel reviews with text average 3.84 stars; the hotels' Google ratings average 4.27
Average star rating of the written reviews we collected, against the overall rating Google shows for the same places (which counts every rating, with or without text), weighted by each place's review count.
Overall Google rating, all ratingsAverage of written reviews, Jan 2025 to mid-2026
Part of that gap is time, because Google's rating reaches back years and the written reviews cover January 2025 to mid-2026. Both point the same way for a reader, though: the reviews you open are not a sample of the average customer, so they should be read by star level, not as a poll.
The gyms show the skew can run the other way. Their written reviews average 4.64 stars, above the 4.53 overall rating, and the written reviews come in lower at only 19 of the 33 clubs. Our report on those clubs found about 4,903 five-star reviews arriving above the normal rate between February and October 2025, with no change in unhappy reviews. When a business asks happy customers to write, they do, mostly in a few words. Who writes depends on who is asked, which is why it pays to track when review requests go out and read each period's reviews against that.
How much more do unhappy customers write?
Several times more. The median one-star hotel review runs 86 words; the median five-star review, 30. At the gyms it is 49 against 11, and on the Duolingo App Store 26 against 9. So the negative reviews take up far more of the text than their share of reviews.
Dimension Labs analysis
One- and two-star reviews are a minority of reviews and a much larger share of the words
Share of written reviews that are one or two stars, and share of all words written that come from those reviews.
Share of reviewsShare of words written
At the gyms, one- and two-star reviews are 7.5% of reviews and 23.3% of the words, about three times their share. Anyone who reads reviews by scrolling, or any summary that counts words or sentences, meets the angry customers first. That is also why reviews are so useful: an unhappy customer explains what went wrong in detail, which a star never does.
Our written reviews are mostly negative but our average rating is 4.1. Which one should I trust?
Both, for different questions. The rating, which counts every customer who scored you, is the better measure of how you are doing overall. The text is the better source of why. The mistake is to read the text as if it were the rating: to count the share of reviews that complain about delivery and present it as the share of customers who had a delivery problem.
Four steps keep the two in their place:
Report the level from all ratings. Use the full rating count and average from the platform, or from your own post-purchase prompt, not the average of the reviews you exported.
Count topics within each star level. Work out what share of one-star reviews mention a topic, what share of two-star reviews, and so on.
Weight back to the full mix. Multiply each star level's topic rate by that level's share of all ratings, including ratings without text, and add them up. A topic named in most one-star reviews matters less if one-star ratings are rare.
Hold review length fixed when you measure what drives bad reviews. Long reviews name more topics and are more often negative, so a topic can look harmful just because it appears in long reviews.
The length effect is large. Among the hotel reviews, 7.0% of reviews of ten words or fewer are one or two stars, against 44.5% of reviews over 100 words. Our causal model for the hotels holds review length fixed for that reason, along with the hotel, tier, language and reviewer type.
Dimension Labs analysis
The longer the hotel review, the more likely it is one or two stars
Share of one- and two-star reviews among 8,504 rated hotel reviews, by review length.
Once length and star level are accounted for, the written reviews say clearly which problems turn a stay into a bad one. In the hotel data, the most-mentioned topic barely moved the chance of a bad review, while cleanliness and staff did; which complaints cause bad reviews sets out the hotel, gym and café results.
Are written reviews a reliable read of customers at all?
Yes, for reasons and for comparisons, as long as you know the tilt. Surveys tilt too: a low-response survey hears from the people who chose to answer. Can reviews stand in for surveys when response rates are low? covers the research on how reviews skew and how to calibrate them against a small survey.
Three habits make written reviews dependable:
Read every review, not a sample of the latest or the longest. Short positive reviews carry praise you will want to protect, such as a named employee or a product feature.
Keep the source and the date on every review. App store, Google, Trustpilot and Reddit each attract different moods, and a review drive changes who writes in a given month. For a rating that is falling, why is our rating dropping shows how to check whether the reviewers changed.
Add the channels where happy customers do talk. Support chats, survey comments and post-purchase prompts reach people who would never write a public review.
Why read every written review with Dimension Labs instead of skimming the angry ones?
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo.
It is fully omni-channel: Google, App Store, Google Play, Trustpilot and Reddit reviews sit next to survey comments, chats and tickets, each with its source kept, so the quiet, happy customers in other channels count too.
It integrates with the data that gives reviews their weight: rating counts, locations, products and order history load as files or through connectors. Joining what customers wrote to revenue and retention is on Prove ($2,999 a month) and up.
It enriches every review, the three-word five-star ones included, with the same Dimensions written for your product or venue, so topic rates can be computed within each star level and weighted back.
Its visual analytics layer turns the result into a dashboard, a monthly deck or a PDF for the product and operations teams, shared by link.
The NBA uses it to read fan feedback at full volume: it analyzes 100,000+ open-ended responses within hours instead of months, and its insight review cycles fell from about 60 days to under 6 hours.
FAQ
Do happy customers write reviews?
Fewer of them, and they write less. In the hotel reviews the median five-star review was 30 words against 86 for a one-star review, and the written reviews averaged 3.84 stars against overall ratings of 4.27. When a business asks happy customers to review, they do write, as the gym data shows.
Why are one-star reviews longer than five-star reviews?
A customer with a problem has a story to tell and often wants a response. A satisfied customer usually has one thing to say. That makes negative reviews the richest source of what went wrong, as long as their share of the text is not mistaken for their share of customers.
What is the best way to analyze written reviews without the negative skew?
For a team that wants every review and every other channel read with one set of Dimensions, topic rates weighted by star level and reports built for each team, Dimension Labs. For a quick read of one export, a spreadsheet with star level as a column and a topic count per level does the core of it.
Should I remove short reviews before analyzing them?
No. Short reviews are mostly positive, and dropping them makes the text skew even more negative. Keep them, and drop only reviews with no text at all from the text analysis while keeping their stars in the rating.
See review analytics in Dimension Labs, or start with one export of your reviews. Plans start at $299 a month with a 14-day free trial.