Search for a free App Store keyword tool and every result promises the same two columns: search volume and difficulty. The first column does not exist. Apple publishes no search-volume figure for any term, in any storefront, to anyone — so whatever number sits in that column was modelled by the vendor, and the free tier is where the model is least likely to be explained.
That is not an argument against free tools. It is an argument for knowing what one can read. Two things are genuinely public: the suggestions Apple's own search box shows as you type, and the live result page for any query. A tool built on those can tell you what people type, how those phrasings order against each other, and how entrenched the apps already ranking are. It cannot tell you how many people type anything. What follows is the method that fits those limits.
What Apple indexes
Everything defensible about what a keyword can do starts with one sentence on Apple's App Store search page, fetched for this post on 9 September 2026:
"Search results are based on a number of factors, including text relevance (matches for your app's title, subtitle, keywords, and primary category), as well as user behavior (downloads, ratings and reviews, and more)."
Four text fields: title, subtitle, keyword field, primary category. The same page rules one field out by name — promotional text "doesn't affect your app's search ranking" — and does not mention the description. Apple says "including", so that is an absence of documentation rather than an exclusion, but a keyword that only fits in the description has no documented home.
The caps come from App Store Connect's reference pages, also fetched today: the name is "no more than 30 characters", the subtitle "can't be longer than 30 characters", and the keyword field is where Apple contradicts itself. The platform version information reference says "You can provide up to 100 bytes of content"; the search page says "Keywords are limited to 100 characters total". In English the two agree. In Japanese, Greek or Arabic a character is two to four bytes and they do not, so plan against 100 bytes. The rest of that field's rules — no repeats across the three fields, plurals count as duplicates, no spaces after commas — are a separate post.
For research the budget is what matters: 30 + 30 + 100, no word repeated across them. Keyword research on the App Store is the job of choosing roughly 160 characters, so a short list you have checked beats a long list you have not.
"Traffic" is a position, not a count
Apple documents that search hints exist:
"Search hints appear once someone inputs text into the search bar and can help inspire people before they search or select a result."
On the same page it says popular terms "may drive a lot of traffic, but are highly competitive", while "Less common terms drive lower traffic, but are less competitive." Apple uses the word traffic and attaches no figure to it, for any term, anywhere.
What Apple does not document is how the hint list is ordered. Everything from here is this site's reading: a suggestion nearer the top of the list is a phrasing more people type. That reading is the basis of every "traffic" number here, and it is why the number is ordinal. The arithmetic is simple enough to print — position 1 scores 100, each position down costs 8.5, nothing scores below 15:
Three consequences follow.
It is not a volume. Position 1 is not "twice" position 5, and 100 counts nothing. Two terms can be compared in one storefront on one day; a term in the US hint list cannot be compared with one in the German list.
A term Apple never suggests has no reading. Not a low reading — none. The tool returns null and shows "unknown" rather than a mid-range 50, because a fabricated 50 is indistinguishable in a table from a measured one.
The hint list is mostly brands. Measured live on 23 August 2026 in the US storefront, the seed habit returns habitica, habitify, habitshare, habitkit and habitlink among its suggestions — real things people type, useless to you unless you own one. The obvious filter, "does the top app's name match the query", fails: developers name apps after generic queries, and it classified habit tracker and invoice maker as brands. What works, and what the app idea finder runs, is the share of the top 10 whose title carries the query's last meaningful word: invoice 1.00, sleep tracker 0.89, habit tracker 0.78, against habitica 0.22 and sleeper 0.10, with the threshold at 0.4. It measures whether the market uses that word for that thing, and it has a cost — invoice generator scored 0.11 because the result set says "maker", so a real query is dropped. That is the safe direction, and every dropped query is listed with its score. In the keyword tool itself you make the same call by eye, from the ten apps shown behind each row.
The hints endpoint answers for 30 storefronts here. Any other storefront says "unavailable" rather than inventing a number, and so does a request that fails — a fact about the network, not about the term.
Difficulty is read off the live result page
The second public thing is the result page: Apple's search endpoint returns the apps ranking for any query, with each app's rating count. That is the material for a difficulty reading, and the formula is worth spelling out because "difficulty" usually arrives as a bare number with a colour:
- Take the top 10 and the log of each app's lifetime rating count — log-scaled, because an app with a million ratings is a different problem from one with a thousand, and a linear average would let one giant swamp the row.
- Weight the top 3 at 1.5 times the rest. Those are the apps you would have to displace.
- Scale so a row where every slot is held by multi-million-rating apps reads 100.
- Add 1.5 per app whose title contains the exact query. A row where everyone already carries the phrase is contested on metadata too.
- Scale down for a thin page. Fewer than 10 results is easy whoever is in them.
The keyword tools and the Apple Search Ads planner share this one formula, so "hard" means one thing across them. Read what it measures: rating mass and title saturation in the top 10. Not quality, not retention — a row of large, badly reviewed apps still scores hard. The tool shows the ten apps behind the number, because a score without the apps that produced it cannot be audited.
The labels — Easy below 35, Medium, Hard from 55, Very hard from 75 — are this site's bands, chosen for readability. Nothing about them is Apple's, and "Easy" describes the incumbents, not your odds.
What no free tool can show you, and no paid one either
A competitor's keyword field. Re-verified on 20 August 2026 against Apple's public API across 133 result objects from six search terms: the keys subtitle, keywords and promotionalText do not exist in the response. Not empty — absent. A tool offering a rival's keyword field is showing an inference from where that app ranks — honest only when labelled as such.
Search volume. Several free tools advertise a per-keyword volume. Apple has published none for them to draw on, so any such figure is a model, and one without a stated method is a guess.
Field weights. The received wisdom is that the name outranks the subtitle which outranks the keyword field. Apple names the four fields and never their order or weights. It may be right; it is practice, not documentation.
Whether an edit worked. Rank moved after a metadata change is an observation. Rank moved because of the change is an attribution, and a competitor's release or an editorial placement moves the same number.
The method, step by step
This is the sequence the free keyword tool is built around, in any of the storefronts the hints cover.
- Seed with what a stranger would type, not your name. "habit tracker", "invoice", "sleep sounds". Your brand is a term you already own.
- Read the hint list as a list. Each row's traffic index is its position. The tool scores the first five automatically so the page is useful before you touch it.
- Expand once. The tool re-queries the seed followed by each of fifteen letters (
habit a,habit b, …) and keeps every suggestion that still contains the seed. Long-tail phrasings a bare seed never shows surface here. - Score the rest, four at a time. Each score is one live search request, so the fan-out is bounded. A failed request is marked failed and retryable, never scored as easy.
- Drop the brands, from the ten apps behind each row; if they are one company's family, the term is theirs.
- Check where you rank today. The rank checker looks 200 deep — as far as Apple's search endpoint returns — in any of 175+ storefronts. "Not ranked" means not in the top 200.
- Write the 160 characters, then check rank again for the same terms and read the result as an observation.
| Signal | Where it comes from | What it measures | What it must not be read as |
|---|---|---|---|
| Traffic index | Position in Apple's search hints | Which phrasings Apple orders ahead of which, in one storefront | Searches, impressions, a percentage, or a figure comparable across storefronts |
| Unknown traffic | A term Apple never suggests | No reading | Low volume |
| Difficulty | Live top-10 rating counts and exact-title matches | How entrenched the incumbents are | Quality of those apps, or your odds of ranking |
| Head-noun share | Titles in the live top 10 | Whether the market uses the word for the thing | A brand detector; it drops real queries |
| Rank | Position in a 200-deep live result | Where you sit today, in one storefront | Where you will sit after an edit |
| Competitor keyword field | — | Not in Apple's public API | Anything a tool claims to show you |
What this cannot tell you
- Not how many people search anything. No source has it, and a tool that prints one has modelled it.
- Not how Apple orders its hints. Apple documents that hints exist and nothing about their order. Position-as-popularity is this site's reading, stated as such.
- Not the weight of any field. Four fields are named; their relative importance is not.
- Not whether an "easy" term is worth having. Difficulty describes the incumbents. Whether anyone typing the phrase wants your app is a judgement the result page cannot make.
- Not the effect of your edit. Correlation between a metadata change and a rank move is an observation, not a cause.
The work
Put a seed into the free App Store keyword tool, score the list, drop the brands, and check your rank for the handful of terms left before you spend any of the 160 characters on them. The bare seed expander on keyword research is the same workbench without the surrounding copy.