Try it with your own rules
Start from the preset below, adapt the questions, and inspect the typed answers. Run it live in the playground — it is the same request shape your application will send.
What Jev returns
Illustrative output for the preset below — run it live to get real values for your input.
Sound quality genuinely surprised me and the battery easily lasts a week. The case hinge feels flimsy though, and the earbuds drop connection about once a day. For this price I would still buy them again.
How many stars does this review imply?
Would the reviewer recommend buying this?
P(yes) = 0.88 — positive overall despite two complaints
Does this read like a genuine purchase experience?
P(yes) = 0.93 — specific pros and cons from real usage
From one example to a reusable workflow
Define the scale in words
Give every level a plain-language meaning ("3 · Mixed", "4 · Good"). The score lands between levels, so you can keep the decimal for sorting.
Compare implied and stated stars
A five-star rating with a text that reads like a two is a data-quality signal. Flag large gaps instead of trusting either number.
Treat authenticity as a triage signal
A low authenticity probability sends a review to a person; it is not proof of fraud. Look at the pattern across reviews before acting on one.
Keep the criteria separate
| Check | What it measures | How to use it |
|---|---|---|
| Stars · Score | What rating does the text imply? | Use the decimal to rank, and the nearest level to display. |
| Recommend · Yes/No | Would they buy again? | Often more useful than stars for product decisions — a three-star review can still recommend. |
| Authentic · Yes/No | Does it read like real usage? | Set a low threshold for review, not for removal. |
The problem
Star ratings are coarse and text is unstructured. Product, support and trust teams want a consistent score and a few yes/no signals they can filter and aggregate.
Aggregate before you act
One review is noisy. The value comes from averaging implied ratings per product or version and watching the authenticity rate over time.
Run it on the whole backlog
A review is short, so a three-question request is a few hundred input tokens. Scoring an entire back catalogue is practical and cheap.
FAQ
Can it detect fake reviews with certainty?
No. The authenticity probability is a signal for triage. Combine it with account age, timing and duplication checks before you remove anything.
Does it explain why a review got a score?
No. Jev returns the decision and its probabilities, not a written explanation. Keep the review text beside the score so a person can read both.
Can I score reviews in bulk?
Yes. Use the Batch tool in the dashboard, or loop over your reviews with the API and keep each request small.