A spreadsheet tells you what happened. This tells you what is wrong.
Exporting reviews gives you rows. Insights reads those rows and answers the question you opened the page for: which item, which criterion, which month. It runs on your own machine, on the reviews you already extracted, and it stays quiet when it has nothing solid to say.
Six things, all derived from real columns.
Nothing here is estimated. Each calculation reads a field the platform actually publishes, which is why what you get depends on where you exported from.
The one line that matters
Every analysis opens with a single sentence: the rating is falling, this item is the weak spot, this criterion decouples, or nothing stands out. It is chosen from the numbers, never written to fill the space.
Rating spread and monthly trend
The distribution from five stars to one, and the average month by month, so a decline shows up as a decline instead of a feeling.
Breakdown by segment
Average and volume per item, per company size, per industry or per location, depending on what the platform publishes. Lowest first, because that is what you came for.
The words that separate
Not the most frequent words, which are the same on both sides, but the ones that appear in the complaints and not in the praise. That is the difference between "delivery" and "long wait".
Sub-ratings and reply coverage
Where the platform scores several criteria, each one is averaged and the one that decouples is named. Where replies exist, the share of reviews that received one is measured.
A report you can send
One standalone HTML file, no external resources, opens offline. Aggregates only: no reviewer names, no review text.
What this is not
It is not a language model. The analysis counts, groups and compares; it never writes a sentence about a review it has not measured. That is a real limitation: a model would read irony and nuance better than any word-frequency method. It is also the reason this can run entirely on your machine, cost nothing per analysis, and never state a finding it cannot show you the numbers for.
It also refuses to speak below a threshold, on purpose. A segment with fewer than five reviews is not displayed. A single month is not a trend. A reply rate of zero is not shown at all, because we cannot tell "this seller never replies" apart from "this platform does not publish replies", and accusing a seller of ignoring customers on that basis would be worse than saying nothing. On a small extraction you will often read "nothing stands out". That is the honest answer, not a failure.
Why it runs in your browser
Reviews are written by named people. Sending them to a server to be analysed would make us responsible for other people's personal data, and would mean a copy of every competitor you study sitting in someone else's database. Computing locally removes both problems, and it is what lets the analysis cost nothing per run rather than a fee per thousand reviews.
The practical consequence is simple: the extraction, the analysis and the exported report all happen on your machine. Our backend only ever sees your account email, your credit balance and billing identifiers.
What the analysis can read depends on the source.
G2 publishes who is speaking, so it supports segmentation by company size. Capterra scores four criteria, so it can name the one that decouples. Etsy attaches the item purchased, so it can name the listing dragging a shop down. This is not marketing: it is the columns.
| Platform | What the analysis can tell you |
|---|---|
| Trustpilot | Month-by-month curve, so a drop shows up as a drop and not as a feeling. |
| Google Maps | The complaints that come back, in the reviewers' own words rather than a summary. |
| Booking & TripAdvisor | The words that separate a good stay from a bad one, taken from the guests' own positive and negative points. |
| TikTok Shop | Average and volume per variant, lowest first. |
| Amazon | Verified versus unverified: whether the good scores come from real purchases. |
| Yelp | The recurring complaints, in the customers' words. |
| G2 | Average per company size: does this product disappoint enterprises and delight small teams? |
| Capterra | The four criteria averaged separately, weakest first. |
| Etsy | Average and volume per item, lowest first. |
| AliExpress | Average per variant, so you order the version that holds up. |
| Wayfair | Average per variant, lowest first. |
Review analysis, answered.
How does the analysis work?
It runs entirely in your browser, on the reviews you just extracted. It reads the columns the platform actually publishes (ratings, dates, sub-ratings, the item or company behind each review) and compares them: distribution, trend by month, averages per segment, and the terms that separate positive reviews from negative ones. No review text is uploaded anywhere.
Is this AI?
No, and that is deliberate. The analysis is statistical: it counts, groups and compares. It never generates a sentence about a review it has not measured, which means it cannot invent a finding. A language model would read nuance better, but it would also have to send your reviews somewhere and it would sometimes be confidently wrong.
How many reviews do I need before it says anything useful?
The analysis deliberately stays quiet below a threshold. A segment with fewer than five reviews is not shown, a single month is not drawn as a trend, and a sub-rating is only called weak when the gap is wide enough to mean something. On a small extraction you will often see "nothing stands out", which is an honest answer rather than a filler.
Can I share the analysis with someone who does not use the extension?
Yes. Insights exports a standalone HTML report: one file, no external resources, opens offline in any browser. It contains aggregates only, no reviewer names and no review text, so you can send it without redistributing anyone else’s reviews.
Does the analysis cost credits?
No. Credits are spent on exported rows. The analysis and the report read data you already extracted, so they cost nothing extra. Insights is a separate monthly plan at $14.99, and it is included in Agency.