Real review distributions are J-shaped: a pile of fives, a smaller pile of ones, and very little in between. The default mix here reproduces that, so an average-rating calculation and a rating histogram both behave the way they will in production.
What you can control
- Four sentiment modes, including a polarised setting that produces only one- and five-star reviews.
- Review bodies match their rating — a two-star review complains, a five-star review does not.
- Helpful-vote counts are heavily skewed, so one review dominates and the rest have almost none.
- Verified-purchase flags appear on about seventy per cent of rows, which is where the badge logic gets tested.
Testing ideas
- Rating distributions. Use the polarised mode (only 1 and 5 stars) to test average and histogram calculations, which real review data often looks like.
- Empty states. Generate zero reviews, then one, then hundreds, and check the summary component at each.
- Verified badges and votes. Turn on verified purchase and helpful votes to test sorting by helpfulness.
- Never publish them. Generated reviews are for testing review components. Publishing fake reviews as real is illegal in many places, including under the FTC's 2024 rule in the US.
What this is not
These are obviously synthetic on close reading. Publishing generated reviews as genuine customer feedback is deceptive and illegal in many jurisdictions — keep them in staging.