The hard part is not finding words. It is telling a bad one from one you judged too early.
Most keyword lists fail the same way: they are written by the person who built the app, in the words that person uses. You call it a habit tracker; the people looking for it type "streak app" and "did i take my pills". Those are different words, and only one set of them is being searched.
Fewer than people expect, and the arithmetic says why. A keyword needs roughly fifty taps before its conversion rate means anything. Your daily budget divided across your keyword list, times a fortnight, has to clear that for each one.
| Budget | At EUR 0.50 a tap | Keywords that can be judged in a fortnight |
|---|---|---|
| EUR 10/day | 280 taps | about 5 |
| EUR 25/day | 700 taps | about 14 |
| EUR 50/day | 1,400 taps | about 28 |
Forty keywords on ten euros a day is not forty experiments. It is forty things you will never know the answer to. More on what that budget buys.
Apple scores search terms from 0 to 100. High-scoring terms are contested by everyone, so a small bid on one buys almost no impressions. Low-scoring specific phrases — "shared grocery list for couples" — are cheap, convert far better, and individually bring almost no traffic. A working list holds both: a few head terms you are willing to pay for, and a tail that quietly does the converting.
This is where most money is lost, in both directions.
A keyword with twelve taps and no installs looks like a failure and is not evidence of one: at a true conversion rate of one in four, zero from twelve happens by chance often enough to be unremarkable. Pausing it is a coin toss wearing the costume of a decision.
A keyword with sixty taps and no installs has answered. It is not still learning. Keeping it because it "might come good" is how a list quietly bleeds.
The threshold is not a feeling. It is whether enough has happened for the number to mean anything, and that is arithmetic you can do.
Adding keywords is the fun part; the money is in the subtractions. Every term your exact-match campaigns own should be blocked in Search Match, or your two campaigns bid against each other for the same search. And searches that are clearly for a different kind of app should be blocked outright. That is what negative keywords are for, and almost nobody uses enough of them.
As many as your budget can give about fifty taps each inside a fortnight, which for ten euros a day at fifty cents a tap is roughly five. More keywords than that does not buy more experiments, it buys the same money spread too thin for any of them to answer.
A few, if you can afford them. High scores mean heavy competition, so a small bid buys almost no impressions. Specific low-volume phrases are cheaper and convert better, and a working list holds both.
You cannot see their keyword lists, and any tool claiming otherwise is estimating. What you can see is your own search terms report, which tells you the exact phrases real people typed before they saw your app — better data than a guess about somebody else.
When enough taps have happened for the absence of installs to mean something — around fifty. Below roughly a dozen taps, zero installs is normal chance and pausing is guessing. Above sixty, it has answered and keeping it is just paying to be told again.