ReviewFold
Guide · Choosing the output
App Review Sentiment Analysis vs Theme Analysis
A star rating says how someone felt when they finished. A theme says which task failed. Those are different products of the same sentence, and mixing them up is how teams end up fixing the loudest thing rather than the most specific one.
“…maybe add voice memos to the conversation for DM’s? My friends and I sometimes move over to Instagram to use the voice memos over there.”
01 · Four layers of one sentence
Star, attitude, task, theme.
They arrive in the same review and answer different questions. Reading all four is what stops a rating from being treated as a finding.
-
Star
The rating the person chose. Comparable across reviews and almost silent about cause.
Tells you: how it ended for them. -
Attitude
What the text sounds like — warm, flat, exasperated. A reader’s judgement, not a measurement.
Tells you: how to hear the sentence. -
Task
What the person was trying to do. Usually one clause, often buried mid-review.
Tells you: what to go and check. -
Theme
A named group of reviews describing the same task. A container you can re-open, not a verdict.
Tells you: what recurs, and where.
Sentiment describes expressed attitude. Themes help identify what that attitude concerns. Use both when you need to understand how people feel about a particular task.
02 · The same five reviews, four ways
A five-star can hold the sharpest complaint.
Five excerpts from three saved samples — Threads, DramaBox and Remini, September 2026. Each row keeps its store, date and rating. The attitude column is an editor’s reading and is labelled as such.
-
Threads ·
“…maybe add voice memos to the conversation for DM’s? My friends and I sometimes move over to Instagram to use the voice memos over there.”
- Star
- 5/5Rated 5 out of 5
- Attitude
- Positivewith a request
- Task
- Send a voice note inside a DM
- Theme
- Message formats
-
Threads ·
“I wish the three dots on the right next to each post had a bigger touch target. It always opens the thread rather than accessing the menu.”
- Star
- 5/5Rated 5 out of 5
- Attitude
- Positiveissue already fixed
- Task
- Open one post’s menu without opening the post
- Theme
- Touch targets
-
DramaBox ·
“I love the dramas but I can’t finish them because I can’t earn coins by watching ads.”
- Star
- 3/5Rated 3 out of 5
- Attitude
- Mixedlikes the product, blocked by the mechanic
- Task
- Finish a series already started
- Theme
- Coin economy
-
Threads ·
“Does not work properly on iPads.”
- Star
- 1/5Rated 1 out of 5
- Attitude
- Negativeno detail offered
- Task
- Not stated — only a device
- Theme
- Cannot be assigned yet
-
Remini ·
“Horrible with ads. When u upload the pic. Ads everywhere im done I’ll find one that doesn’t make me play a game to get this done.”
- Star
- 1/5Rated 1 out of 5
- Attitude
- Negativeangry and leaving
- Task
- Edit one photo without playing something first
- Theme
- Ads before the task
The fourth row is the useful failure. A one-star rating with no task behind it is a signal that something happened, not a theme — and treating it as one would put a guess into your backlog with a date attached.
03 · Matching output to question
Start from the decision, not the technique.
Each output is cheap in a different currency. Pick the one whose blind spot you can live with.
| If you need to decide | The output that answers it | What it still cannot tell you |
|---|---|---|
| How ratings in the collected sample change | Star distribution, tracked over time | Why, and who. A rating is the last second of a longer experience. |
| Which problems keep recurring | Themes with their member reviews listed | How many users are affected, or how severe it is for them. |
| What to investigate next | A task statement per theme, plus the review IDs behind it | Whether a fix works. That needs a test, not more reading. |
| Whether one sentence is positive | A text-based attitude label, checked in context | The specific task, cause and scope of the experience. |
Star ratings and text sentiment can disagree. Neither replaces reading the task and conditions behind a theme; this page provides editorial labels, not a tested sentiment model.
04 · What a theme is worth
A theme is a container. Its value is the reviews inside it.
Two themes can have the same name and mean different things. “Paywall” might hold four reviews about a price, three about a trial that charged early, and two about a screen that could not be dismissed. Counted together they look like one large problem; read separately they are three unrelated ones.
That is why a usable theme has to keep its members: the count, the dates, the ratings, and at least one sentence you can re-read. Without them, the theme is a label someone applied, and the only thing you can do with it is trust it.
Choosing an order after that is a separate job, and it is not decided by theme size. The guide to prioritising review feedback works through why the biggest group is often not the first thing to fix.
05 · Keep this beside the table
Five questions to ask of any review output.
- Is this a rating, a reading, or a task?
If the cell mixes two of them, it will be quoted as if it were the third.
- Whose reading is the attitude label?
Record how the label was assigned. Human and automated interpretations both need checking against the text.
- Does every theme name a task?
“Bugs” and “UX” name categories. “Could not finish an episode without paying” names something you can test.
- Where is the review that disagrees?
A contrasting review can help locate the boundary of a claim. It does not automatically outweigh the supporting accounts.
- What would change my mind?
Write the answer down before you act. If nothing would, the analysis was decoration.
Boundaries
What this page is not claiming.
Four layers are a way to read. They are not a measurement model.
- No sentiment scoreReviewFold does not output a per-review mood value, and this page does not pretend one exists.
- No accuracy claimThere is no benchmark here for attitude labelling, because the labels are editorial interpretations.
- No orderingTheme size does not set priority. That argument belongs to the prioritisation guide.
- No current-state claimThese are September 2026 reviews. A fix since then would not appear.
Current support, stated plainly
- Apple App Store: US and Canada. Google Play: US.
- Up to 200 recent public reviews per report.
- One app and one region per report; no cross-app comparison.
- Reports group feedback into themes with excerpts attached; there is no sentiment score.
- No severity scoring, no automatic priority order.
Themes, with the reviews attached
Get the groups, keep the sentences.
A report names the themes in a recent sample and keeps the original reviews beside them, so you can check a task before it becomes a ticket.