A star rating is a single number asked to summarize a multi-dimensional experience, and the academic record is clear that it often fails at that job. A large-scale study of tourism reviews found an 18.6% rate of rating-text incongruence, and research on app stores puts text/rating mismatches at roughly 20%. In iGaming specifically, public Trustpilot distributions for casino brands are frequently bimodal — large 5-star and 1-star blocks with almost nothing in the middle — which means the headline average conceals two very different populations of players. Our algorithm treats the written text, not the submitted star count, as the primary sentiment signal.
Key takeaways
- Peer-reviewed studies report rating-text mismatches in roughly 18–20% of online reviews across tourism, app-store and hotel datasets.
- Star ratings inflate over time independently of actual satisfaction: research on labor marketplaces found the share of workers with a perfect rating grew from 33% to 85% in six years, while text-derived sentiment stayed flat.
- Casino and sportsbook review pages on Trustpilot often show J-shaped or bimodal star distributions, not the smooth bell curve a single “average rating” implies.
- A 4- or 5-star review can still contain specific, text-only complaints about withdrawal delays, KYC friction or RTP perception that a numeric score never surfaces.
- GamblScout’s algorithm extracts sentiment at the aspect level and flags divergence between the stated score and the text as a distinct signal, not as noise to be averaged away.
Table of contents
The documented gap between stars and text
The idea that a star rating is an unreliable proxy for sentiment is not a fringe claim; it’s an active research area in computational linguistics. A 2026 study using transformer-based sentiment models on more than 16,000 tourism reviews describes the core problem directly:
a review reading “Beautiful gardens, but overpriced and overcrowded” accompanied by a 5-star rating illustrates this tension: the text conveys mixed-to-negative sentiment, while the numerical score signals unambiguous satisfaction
. The same paper found that
the incongruence rate of 18.6% shows that rating-text mismatch is systematic rather than random
, and that
ratings capture overall judgments, while review text captures more specific details of the visitor experience
.
This isn’t limited to tourism. A widely cited analysis of Android app reviews
manually examined 8600 reviews from 10 popular Android apps and found that 20% of the ratings in our dataset were inconsistent with the review
, and a separate LLM-based study of app-store and streaming reviews notes that
approximately 20% of app reviews exhibit inconsistencies that significantly affect user perception and app marketability
. Work on hotel reviews from TripAdvisor adds a nuance relevant to gambling platforms:
on a scale of five stars, those reviews ranked with middle scores include a mixture of positive and negative aspects
— meaning the “3-star” bucket is often where the most genuinely mixed, detail-rich feedback lives, not the extremes.
A 2026 study on a large Italian review corpus summarizes the broader consensus: ratings and text are usually correlated but not perfectly aligned, and
a five-star review may include critical remarks about specific product aspects
. For a sector like online gambling, where the “aspects” in question include withdrawal speed, KYC friction and perceived game tightness, that gap matters more than it does for, say, a coffee grinder.
Why ratings inflate while text sentiment doesn’t
Beyond the text/rating mismatch at the individual review level, there’s a second, compounding problem: aggregate star ratings drift upward over time independent of whether satisfaction is actually improving. Research published in Marketing Science and covered by INFORMS describes this as “reputation inflation.” The study’s lead author summarized the finding bluntly:
“Our research shows across a variety of online marketplaces that five-star feedback increases over time, and it’s not because people are more satisfied with what they receive.”
The same body of research found that
the share of workers with a perfect 5-star rating on an online marketplace for labor grew from 33% to 85% in only six years
, and that
the majority of Uber drivers have a perfect five star-rating
.
A separate study of Vivino wine ratings gives the clearest demonstration of why this matters for sentiment analysis specifically:
the wine platform displays strongly increasing numerical feedback over the time period from 2014 to 2020 while the scores predicted by reviewers’ written feedback remain constant
. In other words, the words people wrote didn’t get more positive — only the number attached to them did. The same research notes that
almost 95% of Airbnb properties are rated at 4.5 or 5 stars
and
almost 90% of UberX trips are rated at the maximum of five stars
, compressing the scale until it stops discriminating between genuinely different experiences.
A 2026 marketplace study adds a mechanism: ratings inflate partly because platform growth changes who is doing the rating.
Onboarding attracts more first-time and non-dedicated reviewers while reducing the share of dedicated, experienced reviewers, and because first-time reviewers tend to give higher ratings than dedicated reviewers, this compositional shift produces aggregate rating inflation.
New, inflated reviews are also measurably thinner:
new reviews are less informative: they are shorter, contain fewer photos, and are written with less effort
. For an industry hub tracking operator trust over time — see our macro-economics coverage — this means a rising average star score can just as easily reflect a changing reviewer mix as a changing product.
What a 4-star iGaming review actually says
Casino and sportsbook review pages rarely show the smooth, bell-shaped distribution a single “average rating” implies. Public Trustpilot data illustrates the point. TheOnlineCasino.com, for example, carries an overall “Average” label with a
TrustScore 3.4 out of 5
built from a distribution where
5-star reviews make up 46% of the total and 1-star reviews make up 31%
, leaving the 2-, 3- and 4-star bands combined at under a quarter of all reviews. online.casino shows a similar split, with a
TrustScore 3 out of 5
built on
33% 5-star reviews against 62% 1-star reviews
, and almost nothing in between. trusted-casinos.top sits at the extreme: a
TrustScore 2 out of 5
, with
93% of reviews at 1 star and 7% at 5 stars
.
| Trustpilot listing | Overall TrustScore | 5-star share | 1-star share | Pattern |
|---|---|---|---|---|
| TheOnlineCasino.com | 3.4 / 5 | 46% | 31% | Bimodal, thin middle |
| online.casino | 3.0 / 5 | 33% | 62% | Negative-skewed bimodal |
| trusted-casinos.top | 2.0 / 5 | 7% | 93% | Near-unanimous negative |
The pattern matches what rating-inflation researchers call a J-shaped distribution:
a highly skewed distribution of online ratings
in which the middle is hollowed out. An average that blends a 46% 5-star block with a 31% 1-star block is not describing a “mostly good, somewhat flawed” operator — it’s describing two populations with fundamentally different experiences, and the single headline number erases that distinction entirely. This is the same structural issue our sentiment analysis methodology was built to address: reading the text of both blocks separately rather than averaging them into one score.
The six mismatch patterns our algorithm watches for
The tourism-review study referenced above didn’t just measure how often ratings and text diverge — it classified the mismatches into recurring, named patterns, which is useful for building detection rules. The researchers found that
the six-pattern structure shows that incongruence is directional rather than accidental, with “Conservative Rater” and “Obligatory 5-Star” patterns dominating the mismatched cases, indicating that reviewers do not simply make random rating errors
. A “Conservative Rater” writes positive text but under-scores out of habit or caution; an “Obligatory 5-Star” writes mixed or mildly negative text but defaults to the maximum score anyway — a pattern platform-design researchers link to social pressure not to penalize a service the reviewer otherwise intends to keep using.
The practical consequence for anyone building or consuming sentiment models is spelled out plainly in the same paper:
for NLP pipelines, this means that star ratings do not function equally across contexts as weak sentiment labels, and using them without validation can introduce systematic label noise into sentiment analysis models.
That is precisely why our fake-review detection work and our shill detector both treat the submitted star score as one weak input among several, rather than as ground truth.
How this changes operator comparisons
Once rating-text divergence is treated as a measurable property rather than statistical noise, it becomes a scoring input in its own right. Aspect-based sentiment analysis — the technique of tagging sentiment to specific topics within a single review rather than scoring the review as a whole — is well established in adjacent research. A 2026 study built specifically on Trustpilot data
extracted fourteen aspects from the literature and empirical analysis of 3500 randomly selected reviews from the Trustpilot platform
, and found that a transformer model
identified as the best-performing model, achieving over 90% accuracy
at classifying sentiment per aspect. Applied to gambling reviews, this means a single 4-star review can simultaneously register as positive on “game variety,” neutral on “customer support,” and sharply negative on “withdrawal speed” — three separate signals a headline star count cannot carry.
For player-facing comparisons, this matters most where star scores cluster near the top of the scale and stop being discriminating, mirroring the inflation pattern already documented for Airbnb and Uber. An operator sitting at a 4.0-star aggregate built mostly from short, low-effort 5-star reviews and a smaller cluster of detailed 1-star complaints about delayed withdrawals is a materially different proposition than one at 4.0 stars built from consistent, moderately positive, detail-rich reviews across the board — even though the two would look identical on a comparison table that only shows the average. This is the logic behind our scoring system and algorithmic weights, which treats review volume, recency and text sentiment as separate weighted inputs rather than collapsing them into a single star figure.
It also intersects with jargon detection. A complaint phrased in sports-betting slang or crypto-casino shorthand can read as neutral to a generic sentiment model while actually describing a serious grievance — a problem addressed in our work on decoding player slang. Misreading slang as neutral has the same net effect as trusting an inflated star score: it understates real dissatisfaction.
Frequently asked questions
Does a 4-star rating ever mean a casino is unreliable?
It can. Research on rating-text mismatch shows that
a five-star review may include critical remarks about specific product aspects
, and the same logic applies one tier down. A 4-star aggregate built on a bimodal distribution — many 5-star and many 1-star reviews with few in between — can hide serious, recurring complaints that a single average score smooths over.
How common is the mismatch between star ratings and review text?
Across different domains, researchers consistently find mismatch rates in the same range.
A manual review of 8,600 Android app reviews found that 20% of ratings were inconsistent with the review text
, and a large tourism dataset found an incongruence rate of roughly 18.6%. There is no evidence that iGaming reviews are immune to this pattern.
Why do online ratings inflate over time?
Multiple mechanisms contribute. Research in Marketing Science attributes it partly to social pressure to rate “above average,” while a 2026 marketplace study found that
platform growth attracts more first-time, non-dedicated reviewers who tend to give higher ratings than experienced reviewers, producing aggregate rating inflation
even without any manipulation by the operator.
Can sentiment analysis replace star ratings entirely?
Not entirely — stars remain a fast, low-friction input players understand instantly. But used alone, they are a weak proxy for the underlying experience. Aspect-based sentiment models that score specific topics within the text, as documented in Trustpilot-based research achieving
over 90% accuracy
, provide a more granular signal that complements rather than replaces the star score.
How does GamblScout weigh text sentiment versus the star score?
Our algorithm does not average star scores in isolation. It extracts sentiment per aspect (payouts, KYC, support, game fairness) from the review text, cross-references it against the submitted star rating, and flags divergence as a distinct data point rather than discarding it as noise, consistent with the methodology described in our sentiment analysis explainer.
Our methodology
For this article, GamblScout.com’s algorithm draws on peer-reviewed research into rating-text incongruence and reputation inflation, combined with publicly available Trustpilot star-distribution data for named operators, sampled directly from each listing. For ongoing operator scoring, the same pipeline extracts aspect-level sentiment from scraped review text independently of the submitted star rating, cross-checks it against our fake-review and shill-detection models, and weights the result inside our broader AI-driven review framework, in place of relying on a single human-assigned score.
Gambling involves risk. Only play with money you can afford to lose and use the deposit limits and self-exclusion tools available in your jurisdiction.
