We've retrained the AI models behind our Semantic Analysis engine, the part of CXP that finds each opinion in a review, tags its sentiment, and sorts it into a subcategory. The update rolled out between October 28 and October 30, 2026, across all 27 languages. You'll see more of what your guests say, sorted into the right subcategory more often.
Benefits of This Update
- See more of your guest feedback, including opinions the previous models missed.
- Rely on a more accurate Sentiment Score.
- Spot issues faster, with more opinions sorted into specific subcategories.
What's Improved
Overall accuracy is up 5.1 percentage points, and every one of the 27 languages improved. French, Thai, and Spanish gained the most. Here's what that looks like in real reviews:
- Mixed opinions stay separate: "Minibar is reasonably priced, but hotel laundry service is more pricey." used to be one neutral opinion. Now it's a positive opinion on Minibar - Price and a negative one on Laundry Service - Price.
- Short reviews get the full picture: "Small, but clean rooms." used to count as one negative opinion. Now it's two: a negative one about the room size and a positive one about cleanliness.
- More specific subcategories: "Need to replace public seating." moved from Shared Facilities - Others to Shared Facilities - Maintenance.
What You Might Notice in CXP
- More opinions: Your reviews may show more opinions than before.
-
Opinions move to better-fitting subcategories: Expect a lot of opinions to shift into more specific, more descriptive subcategories, including many that used to sit under "Others". For example:
- Public Health is retired: The subcategory stays in CXP, but it's reserved for COVID-era health measures, so new comments won't show up there. Comments about your property's medical or health services now go under Health Measures.
- Music counts toward Ambience: Comments about background music now go under Ambience instead of Entertainment.
Your Edits Helped
Every correction you make in CXP helps the engine learn. This update was trained on thousands of them.
For more details, check: Edit Sentiment · The Semantic Analysis Engine