Review X-Ray doesn't summarize reviews the way a general-purpose chatbot does. It runs every review through the same three-step pipeline, so every number in your report is something you can trace back to an actual customer's words.
1. Scan
Open any page with customer reviews — a Trustpilot profile, an App Store listing, an Amazon product page, a G2 or Capterra comparison — and the extension finds the repeated block of review text on that page automatically. You confirm the right block before anything runs.
2. Classify
Every single review is judged individually by a fast, typed decision model: which category it belongs to, whether it's positive, neutral or negative, whether it's an explicit feature request, whether it implies an unmet need even if it doesn't ask for anything directly, and — for negative reviews — how severe the complaint actually is, from a minor annoyance to a dealbreaker.
This step doesn't generate text. It answers narrow, typed questions about each review, which is what makes it fast enough to run on hundreds of reviews and reliable enough to count.
3. Synthesize
Once every review is classified, the counts, percentages and severity breakdowns are computed in plain code — not estimated by a language model. Only after that does a writer model read the pre-computed numbers and turn them into the strengths, weaknesses and opportunities you see in the final report. It's never asked to count anything itself; it's only asked to explain what the numbers already show.
A model that both counts and writes tends to round numbers, miss reviews, or state a percentage that "feels right" rather than one that's actually correct. Separating counting from writing is what keeps a Review X-Ray report accurate at hundreds of reviews, not just at ten.
What you get
The finished report shows a sentiment breakdown, a pain map of every category by frequency and severity, ranked opportunities with the evidence behind each one, real customer quotes, and a one-click PDF export — all without you reading a single review yourself.
Frequently asked
Does it read every review, or just a sample?
It classifies every review it finds, up to a per-run cap for very large batches. When a batch exceeds that cap, Review X-Ray samples evenly across the full set rather than just reading the first page of results, so the report stays representative.
Can it get a review wrong?
Yes — like any automated classification, individual calls can be uncertain. Low-confidence classifications are flagged directly in the report rather than hidden, so you know which conclusions to double-check.