How we analyse guest reviews
On our hotel pages and ranked lists you'll see figures like “92 of 92 mentions of the food are positive”. This page explains exactly how those figures are produced — so you can judge for yourself what they are worth.
Where the reviews come from
The analysis uses a sample of stored reviews from Booking.com and, where available, Google. The sample may include the newest, highest-rated and lowest-rated reviews. Availability varies by hotel and language. The count shown with an analysis refers to the reviews in its input. The larger review counts behind Google and Booking ratings are the sources’ totals; they do not indicate how many texts Tripplo has analysed or how many unique guests are represented.
How mentions are counted
Our software reads the first 500 characters of each review text in the sample and classifies mentions in eight categories: food, rooms, pool and beach, service, location, condition and cleanliness, noise, and value. Mentions are classified as positive, negative or mixed. The program then counts the classifications per review and category. “92 of 92 food mentions were positive” means that all 92 counted food mentions in the sample were classified as positive. It is not a rating from all of the hotel’s guests.
What we publish
A category appears in the hotel page analysis only when it has at least three counted mentions. Our software summarises strengths and complaints, and selects and translates short quotes from the input. Read these as translated excerpts, rather than guaranteed verbatim quotations from the original. Classifications and summaries may contain errors or omit details.
What the analysis date means
The analysis date records when the text analysis was generated, not when the reviews were written or collected. Hotels in a list may have been analysed at different times; the latest analysis date refers to the most recent of those analyses. Guest ratings and source review counts are updated separately and may carry different dates.
Limitations — what these figures are not
The sample is limited and not random. It does not cover all reviews, languages or time periods, and the same guest may appear on multiple sources. Shortened texts may omit important information, and automated classification can misinterpret their content. The proportions describe classified mentions in the input and are not representative measures of every guest’s experience. Hotels do not pay to be included in the analysis.