Do Travellers Rate Sri Lanka - Or the Hotel They Slept In?

Short answer

Sri Lanka's reviews are, almost entirely, hotel reviews. Across the 315,523 reviews with written text in our corpus, hotel-facility words (room, clean, staff, hotel, pool, breakfast, service) make up 4.29% of every word written, while words about the place itself (beach, view, location, nature, temple, landscape, town) make up just 1.76% — a 2.44× hotel skew. Travellers are scoring the bed, the bathroom and the breakfast far more often than the country they flew 8,000 km to see. And that matters, because when a review does talk about the place, the score is just as high — the place is not the problem. It is what is being left unsaid.

Why this question is worth asking

The single most useful thing you can do with a review corpus is not to rank hotels. It is to find out what people are actually judging. If tens of thousands of guests score a country on its air conditioning, then the country's average score is really a hotel-quality score wearing a destination's name — and anyone planning a trip from those numbers is measuring the wrong thing.

So we split the 315,523 Sri Lanka reviews by the vocabulary people used: reviews whose language is dominated by the room and the property, versus reviews whose language is dominated by the beach, the view, the town and the landscape. Then we compared the scores.

Finding 1: the vocabulary is 2.4× more hotel than place

We counted the share of every written word that comes from each family of terms.

• Hotel-property terms: 4.29% of all words (room, rooms, clean, staff, hotel, pool, breakfast, service, bed, bathroom, shower, reception, towels, check-in, wifi, air con)
• Place terms: 1.76% of all words (beach, view, views, location, nature, temple, landscape, scenery, town, village, sunset, sunrise, mountain, sea, ocean, surroundings)

That is a 2.44× skew toward the property. The average Sri Lanka review is a hotel inspection with a holiday attached, not a holiday report with a hotel attached.

Finding 2: but reviews that talk about the place score just as well

We isolated the short, happy reviews — under 300 characters, scored 9 or 10 — because those are the ones that say the one thing that mattered most. Then we split them by whether their language was place-only or hotel-only:

• 7,536 place-only short reviews → mean score 9.71/10
• 40,588 hotel-only short reviews → mean score 9.72/10

The difference is a rounding error. When a guest writes about the beach, the view and the walk, the verdict is just as positive as when they write about the room and the staff. The destination is not dragging scores down — it simply is not being asked about.

This is the same shape as the finding in our 46-Word Rave study, where the view and the staff outranked the pool by three to one. Travellers love the country. They just describe the hotel.

Finding 3: which destinations get talked about as places

Per destination, we measured how many place-mentions a region earns per 1,000 words of its reviews. These are the places guests actually describe as scenery rather than as somewhere to sleep:

• Ella — 24.7 place-mentions per 1,000 words (mean score 8.8, n=25,403)
• Mirissa — 23.2 place-mentions per 1,000 words (mean score 8.53, n=13,423)
• Wadduwa — 22.8 place-mentions per 1,000 words (mean score 8.04, n=872)
• Hikkaduwa — 22.5 place-mentions per 1,000 words (mean score 8.82, n=7,400)
• Haputale — 22.2 place-mentions per 1,000 words (mean score 8.76, n=3,598)
• Matara — 21.7 place-mentions per 1,000 words (mean score 8.68, n=2,457)
• Trincomalee — 21.3 place-mentions per 1,000 words (mean score 8.26, n=9,487)
• Colombo — 21.3 place-mentions per 1,000 words (mean score 8.24, n=29,258)
• Bentota — 21.1 place-mentions per 1,000 words (mean score 8.54, n=7,846)
• Unawatuna — 21.0 place-mentions per 1,000 words (mean score 8.59, n=8,137)

Ella, Mirissa, Hikkaduwa and Haputale top the list: guests write about the tea hills, the surf and the coastline, not the lobby. National parks and resort towns sit at the bottom — a safari camp review is mostly a tent review.

Finding 4: the four biggest places, ranked honestly

Among destinations with 5,000+ text reviews, the mean score ranking is:

Note the pattern. The better-scoring large destinations — Sigiriya (9.02), Tangalle (8.87), Ella (8.80), Galle (8.68) — tend to be the ones whose reviews are about the place. Colombo, the largest sample of all at 29,258 reviews, has the lowest mean of any big destination (8.24) and a mid-table place-focus: it is being scored, above all, as a hotel-and-transit city, and scored accordingly.

Finding 5: where the place beats the hotel

The most actionable cut is the gap. In each destination, we compared the mean score of place-only short reviews against hotel-only short reviews. A positive gap means the place consistently out-scored the property:

Arugam Bay (+0.67), Tissamaharama (+0.60) and Trincomalee (+0.54) are the clearest examples of a place out-performing its accommodation. These are, not coincidentally, places where the thing you came for — the surf, the safari, the east-coast beach — is genuinely world-class while the hotel stock is thinner. If you read a hotel-only review of Arugam Bay and skip the place reviews, you will undersell it badly.

Alone at the other end sits Colombo (−0.25) and Negombo (−0.16): the only big destinations where hotel-only reviews outscore place-only ones. People judge these as transit cities, and the ones who write about the room had a better time than the ones who wrote about the town.

Finding 6: the quiet destinations nobody mentions

One place-word cluster is worth separating out, because it is the thing a certain kind of traveller searches for and almost no site indexes: quiet. We measured how many reviews in each destination contain quiet / peaceful / tranquil / calm / relaxing / serene:

• Kosgoda — 21.5% of reviews use quiet/peaceful/tranquil (mean score 9.05, n=480)
• Kurunegala — 19.7% of reviews use quiet/peaceful/tranquil (mean score 8.59, n=770)
• Beruwala — 18.8% of reviews use quiet/peaceful/tranquil (mean score 9.04, n=457)
• Chilaw — 15.2% of reviews use quiet/peaceful/tranquil (mean score 8.75, n=888)
• Batticaloa — 15.0% of reviews use quiet/peaceful/tranquil (mean score 8.21, n=1,143)
• Bibile — 15.0% of reviews use quiet/peaceful/tranquil (mean score 9.14, n=493)
• Kalpitiya — 14.6% of reviews use quiet/peaceful/tranquil (mean score 8.71, n=2,768)
• Ampara — 14.1% of reviews use quiet/peaceful/tranquil (mean score 9.03, n=482)

Kosgoda (21.5%, mean 9.05), Beruwala (18.8%, mean 9.04) and Kalpitiya (14.6%) are the places guests describe as genuinely quiet — and they score at or near the top of the national table. These are not headline destinations. They are the ones to check when the review you are reading for Ella complains about the noise.

What this means if you are planning a trip

• Read the place reviews, not just the top-rated ones. Sort by the reviews that mention the beach, the view or the walk — they predict whether you will enjoy the trip far better than the room vocabulary that dominates the average.
• A 9.5 in Colombo is not a 9.5 in Arugam Bay. In Colombo, hotels score above the place; in Arugam Bay, the place scores above the hotels. The same number means different things.
• Add a quiet filter. If you want peace, Kosgoda, Beruwala, Kalpitiya and Tangalle are where guests actually say so.
• Trust the destination with the thinner hotel stock if the thing you came for is the place. The gap table above is a list of places whose hotels undersell them.

How this study was built

All figures are computed directly from our corpus of 315,523 Sri Lanka hotel reviews carrying written text (Booking.com-sourced, property-level, spanning every region of the island). Vocabulary shares were computed over the words of each review; a review is classified place-only or hotel-only when it uses terms from exactly one family. Short reviews are text under 300 characters with a score of 9 or higher (162,121 qualify). Per-destination figures use all destinations with at least 400 written reviews. The national mean score across the corpus is 8.609/10. Numbers are rounded; some totals may not sum exactly.

Explore the underlying data in our review statistics, find your own match with the destination matcher, or see who should go where.