Review miningReview mining
What is in this resource:
How to build your ASO strategy from end to end
Learn to pick target keywords, write metadata, follow rank and base every decision on data.
From data to action
Simplify the keyword data and stay in line with App Store rules.
Tools that lift rank
Carry the keywords drawn from competitor reviews into your metadata and take your app out of the cusp band into the top 10.
Why keyword selection is an elimination job
Everything starts with a list. Without knowing which phrases you will track there is nothing to measure. This guide is about widening the pool from five channels and then cutting it down to something workable in a week.
Most teams treat keyword selection as a finding problem: find the right words, write them down, done. The real work is not finding, it is eliminating. Hundreds of phrases can be found for any app; most of them are either not searched, out of reach, or do not describe what the app actually does.
Without elimination what happens is this: a fifty-phrase tracking list is built, nothing is touched, and three months later the list is forgotten. A five-phrase list gets looked at every week and actually worked on.
So this guide has two halves. The first is widening the pool: gathering as many candidates as possible. The second is narrowing it: measuring the candidates and eliminating most of them. They are separate mental tasks and doing them together does both badly.
Do not filter while widening. Leaving a phrase out because it "looks too hard" is deciding without measuring, and decisions made without measuring usually turn out wrong - difficulty is not something intuition estimates well.
The five channels that widen a pool
How wide the pool is determines the quality of the opening you eventually find. A narrow pool does not produce a good list. There are five channels and all five have to be used; most teams use only the first two.
- Your own metadataYour app name, subtitle and description are already a source. The two- and three-word combinations of those words form the base of the pool. The advantage is that relevance is certain; the disadvantage is being trapped inside your own language.
- Rival metadataThe names, subtitles and descriptions of the apps in your category. What comes out shows you positionings that never occurred to you. Choosing rivals matters: the apps most like you but larger than you are the most productive source.
- Search suggestionsThe completions returned when you type a word into the App Store search box. These are genuinely searched phrases, and the popularity signal comes from here too. Scanning from a single letter and going deeper produces a wide candidate list.
- Category chartsApple's category rankings are public. The metadata of the first hundred apps in your category shows the category's shared language. This channel is especially valuable when entering a new category.
- Rival reviewsThe source no keyword tool looks at. Users describe the app in their own words rather than your marketing language, and those words sit written in the reviews. Competition on what comes out of this channel is usually low, because nobody is looking there.
The fine print of review mining
The fifth channel is the most productive and the one needing the most processing. Reviews are made of sentences, keywords of phrases; there is a conversion in between and done carelessly it produces rubbish.
The first problem is pronouns. People write "it really helped me" and "it genuinely surprised me". Those are grammatically sound and frequent, but nobody searches them on the App Store. Phrases containing first- and second-person pronouns are dropped.
The second is the app name itself. A rival's name recurs constantly in its reviews and surfaces as the most frequent phrase. A brand name is a keyword, but a rival's brand is no use to you; those repeats are dropped.
The third is word boundaries. In writing systems that do not separate words with spaces - Chinese, Japanese - where a word ends is not marked. Those languages need a separate boundary-finding step, and without it the extracted phrases are meaningless.
The fourth is letter folding. A phrase is lower-cased on the way into the pool, and the rule depends on the country. In Turkish and Azerbaijani a capital I becomes a dotless ı; everywhere else a dotted i. Apply one rule everywhere and a phrase like "AI" becomes "aı" in the Turkish pool, matching nobody who searches for it.
The fifth is volume. Phrases mined from an app with few reviews stay statistically meaningless. To get value here you have to look at the few apps in the category with the most reviews.
- Drop phrases with pronouns; "it helped me" is never searched.
- Drop rival brand names; they bring you nothing.
- Space-free writing systems need a separate boundary step.
- Letter folding depends on the country; one rule everywhere corrupts the phrase.
- Do not trust phrases mined from apps with low review volume.
Measuring the pool
The pool is wide. Now it gets measured - and the order matters, because each measurement has a price.
Popularity first. A relative value is taken from search suggestions for every phrase. It is a cheap measurement and it eliminates most of the pool at once: where no suggestion returns, the phrase is either not searched or not measurable. Because we do not separate those two, the field is left blank rather than written as zero.
Rank second. Each remaining phrase is searched in that country and the first 200 are read. This one is expensive: every search is a request to Apple and the requests per second are capped. So measurement runs per phrase rather than per row - every app tracking the same phrase is read out of one search.
Difficulty third. Computed from the data in the same search: how many apps carry the phrase in their title, what the rating counts at the top are, how many words the phrase has. It needs no extra request, so it comes free.
After those three every row in the pool carries four facts: the phrase, its popularity, your current rank and its difficulty. The list has become a table, and decisions can be made on a table.
From pool to list: the order of elimination
The table is in front of you. Now you reduce it to something workable in a week. The order below is the one that avoids unnecessary work.
- Set aside the blank popularity rowsNo decision can be made about an unmeasured popularity. Those rows go to a separate list and wait there; they are not deleted, because they may become measurable later.
- Eliminate the irrelevantThe pool grows automatically, so phrases with nothing to do with the product get in. A person has to make this cut: if the answer to "does our app actually do this?" is no, the phrase does not enter the list. Ranking for work you do not do means poor ratings.
- Split into bandsDivide what remains by current rank into three: the top 10 (defend), 11-30 (spend), above 30 and outside 200 (invest). The three need different work and keeping them in one list is confusing.
- Take a few from each bandTwo from the defence band, three from the spending band, one from the investment band. A six-row list is one that can be looked at weekly and actually worked on.
- Repeat per countryThis list is for one country. In another the same phrases will sit in different bands, so the list will differ. Applying one list to every country means working on the wrong thing in most of them.
Phrase length changes the behaviour
Phrases in the pool run from one word to four, and length changes how a phrase behaves from top to bottom.
One-word phrases are the most searched and the hardest. Reaching the first page for "fitness", "budget" or "translate" means competing with the category's largest apps; a new app getting in there with metadata is not practically possible.
Two-word phrases are the balanced zone. They are searched enough and their competition is far lower than one-word phrases. Most of the cusp band is this length, and metadata moves work best here.
Three-word phrases are searched less but their relevance is very high. An app ranking for one is doing exactly what the searcher asked for, and conversion is high. Gathered in clusters they produce serious traffic.
Four words and above is usually a fragment of a sentence and rarely searched. Review mining produces many phrases of that length and most of them should be eliminated.
A practical note: the words in the keyword field combine freely into phrases. Writing "learn" and "english" separately also indexes "learn english". So writing a two-word phrase as separate words gets more phrases out of the same field than writing it whole.
| Length | How it behaves |
|---|---|
| One word | Heavily searched, very hard. Not taken with metadata by a new app. |
| Two words | Balanced. Most of the cusp band; where moves work. |
| Three words | Lightly searched, highly relevant. Gathered in clusters. |
| Four and above | Rarely searched. Review mining produces many; most should go. |
Keeping the list alive
A tracking list is not something built once and left. It starts ageing the day it is built, and a list nobody has looked at for three months no longer describes the app.
The first source of ageing is the product. Adding a feature makes your app relevant to new phrases, and the list does not learn that by itself. Every release deserves the question: what new phrase did this release make possible?
The second is the market. Rivals come and go, new phrases become popular, old ones are forgotten. Re-scanning the pool monthly produces new candidates for the list.
The third is your own success. When you carry a phrase to the first page it moves from the spending band to the defence band and its place in the list changes. Not moving it means working the same phrase again and again.
The fourth is countries. Opening a new storefront needs its own list. Copying the existing one is the most common mistake - the same phrases can sit in completely different bands there.
Our suggestion: look at the list weekly, re-scan the pool monthly, keep country lists separate. Those three habits are enough to keep it alive.
Frequently asked
- How many keywords should I track?
- Keep the pool wide and the list narrow. The pool can hold hundreds of phrases; the list you work on weekly should not exceed six. A fifty-row tracking list ends with nobody touching any of the fifty.
- Are the phrases from rival reviews really searched?
- Some are, some are not. That is why they do not go straight onto the list: they pass the same measurement as everything else. Coming from a review is not a privilege, only a source nobody looks at.
- What if I target a phrase describing something my app does not do?
- You can rank for it, but the arriving user will not find what they came for. The result is low-rated reviews, and the rating score affects your strength across every phrase. You win one and lose all of them.
- Can I use the same list in every country?
- No. The same phrase can sit at 13 in one storefront and 62 in another, which means different work in each. The phrases themselves also differ by country.
- How often should I re-scan the pool?
- Monthly is enough. Scanning more often produces no new candidates; categories do not change on a weekly scale. An extra scan in the months you ship is useful, because when the product changes the relevant phrases change too.
Explore our resources
The turning point in ASO: a guide to raising visibility and staying in the race
Ready to rethink ASO from the start? Our new guide explains step by step the road from rank to download. We show how to find the keywords in the cusp band, which phrases to draw out of competitor reviews and how to build the 100-character field. Download the ASO guide now and pass the app ahead of you in search!
See resource
Metadata optimisation - Part II
When an app falls outside the top 10 on the App Store, it loses most of the visibility that comes from search. The real gain is hidden in the 11-30 cusp band: the words only one step from the first page. This guide explains step by step how to find that band and carry it up.
What is inside?
- A clear frame for how the cusp band is measured.
- The method for drawing keywords out of the user reviews of competitor apps.
- An analysis of the mistakes most often made when building the 100-character keyword field.
See resource
The complete guide to optimising your app title
Most apps lose downloads not because they do not appear in App Store search, but because they appear on the wrong words. Our new report, Finding Keywords with Review Mining: Methods and Application Strategies, explains how keywords are drawn out of the source nobody looks at.
Discover the ways to pick meaningful phrases out of the user reviews of competitor apps, prioritise them by country and place them in the 100-character keyword field. Learn which signals to look at while carrying cusp band keywords into the top 10.
See resource