AI summary and pattern analysis

The AI summary panel sits above your review list and turns whatever reviews are currently in view into a written summary, a set of weighted themes, and a short list of issues that need attention, so you don’t have to read every review yourself.

How it works

The panel is collapsed by default. Expanding it for the first time on a given filter set generates a summary from whatever reviews match your current date range, app selection, and other filters, so narrowing your filters before expanding gives you a more focused summary. While it works, the panel shows a progress indicator and can take up to about 15 seconds. Summaries are cached for a few minutes per filter set, so reopening the panel doesn’t always regenerate. The regenerate icon in the summary footer forces a fresh run. Summarizing needs at least a handful of reviews with text. If your filter set is too small or the app was only just added, the panel returns a short note asking you to check back once more reviews have been processed, instead of a summary.

What you get

An executive summary: a few sentences in plain language describing the overall picture for the filtered set, with the most important phrases bolded. Weighted themes: the biggest patterns in the filtered reviews, each with:

• A share of reviews it represents, shown as a percentage with a bar beside it, for example “25% of reviews”. A review can mention more than one theme, so the shares can add up to more than 100%. When they do, a note under the themes says so.
• A sentiment (positive, negative, or mixed)
• A short description plus one actionable recommendation
• Sub-topic pins showing the specific tags that make up the theme, each with its own review count
• Sample quotes pulled directly from real reviews
• A Sample reviews link that filters the review feed down to just that theme
• A new badge when the theme is a topic the AI just discovered, and an anomaly note when the theme’s volume is unusual for the period

Needs attention: a short list of specific issues pulled out of the themes, each with a severity (high, medium, or low), shown separately from the themes so the things that need a response don’t get buried in the narrative.

Two ways themes get built

If your organization has topics defined and enough of your reviews already carry a topic, the summary builds its themes from your existing topic taxonomy, so the same topic reads consistently from one summary to the next. Without enough topic coverage, it falls back to clustering reviews by their content directly. When coverage is low, the panel nudges you to run topic discovery so future summaries land on the tag-aware path. When the summary discovers genuinely new topics along the way, it says how many and suggests them on your Topics page for review. They are not active until you approve them.

Custom topic search

Type any topic into the Get a review summary for any topic… field and click Set custom topic to get a focused summary of just the reviews about that subject, using semantic search rather than a literal text match. That way “won’t load” and “crashes on startup” can surface the same underlying reviews. For a custom topic, each theme’s share is of the reviews that matched your topic, and says so: for example “25% of 75 matching reviews read”.

Comparing periods

The panel summarizes the selected period only. There is no period-over-period comparison in the review feed; for that, use Compare with the period before on Review analysis.

After you have a summary

The footer shows how much reading time the summary saved you (based on the number of reviews analyzed), and gives you a few quick actions: thumbs up/down feedback, copy to put the summary and its themes on your clipboard as text, and regenerate to bypass the cache and run it again.