I was writing store copy for a self-management app and nearly shipped the line “we’ll cure your three-day resolve.” Before writing it I measured what people actually type, and after measuring I deleted the line.
Korean has a fixed phrase for this, 작심삼일, literally “a resolve that lasts three days.” It is the exact thing every habit app claims to fix. It also turns out to be a good place to look, because a single phrase concentrates the demand that English spreads across a dozen different wordings.
Autocomplete is the honest way to measure demand around habits
Google and Naver only surface a suggestion when real people have searched it, which makes it the most honest read on demand around habits that you can get without a paid tool. The catch is that you learn relative ordering and nothing about absolute volume.
On 2026-08-16 I seeded that one phrase, appended eight suffixes, and ran eight queries. Google returned 36 distinct suggestions and Naver returned 10. Here is how the 36 split by what the searcher seems to want.
| What the search is after | Count |
|---|---|
| Definition (meaning, hanja, origin, “in English”) | 15 |
| Assumes it has already happened again | 5 |
| How to fix it | 3 |
| Why it happens | 3 |
| Apps, planners | 3 |
| Everything else (memes, forums, blogs) | 7 |
The largest group was homework
Fifteen of the 36 are definition lookups: meaning, the Chinese characters, the origin story, example sentences, the antonym, how to say it in English and in Japanese.
That is a student learning an idiom, not a person trying to change their behaviour. Nearly half of this keyword has nothing to do with self-management at all.
I put that first because everything below only holds for the 21 that remain. Treating all 36 as evidence that “people struggle with this” would overstate it by about half.
Searches that assume a repeat outnumbered searches for a cure
Three suggestions ask how to fix it. Three ask why it happens. That was the entire shape I expected before I ran the query.
The remaining group was the interesting one:
- three-day resolve, round N
- three-day resolve, multiple times
- three-day resolve, again and again
- three-day resolve, every morning
- 작심삼일도, a dangling particle meaning “even a three-day resolve…”
All five presuppose that this is not the first attempt. Someone typing “round N” is keeping count of their own attempts. That group is larger than the fix group and larger than the cause group.
The phrase people type is not in the dictionary
That last suggestion, 작심삼일도, ends on a particle meaning “even a three-day resolve…” Naver’s top ten carries it a step further: “even a three-day resolve, if ten times.” The completion everyone reaches for is even a three-day resolve, done ten times, makes twenty days.
The arithmetic is honest. Ten rounds of three days is thirty, but the saying lands on twenty, because it counts the days you stopped.
I was about to call this a proverb, so I checked. It is not one, at least not officially. The national standard dictionary lists exactly one entry for 작심삼일, a noun meaning “a resolve firmly made that does not survive three days, referring to weak determination.” Nothing about ten times, nothing about twenty days.
So the failure reading is the one that got written down, and the cumulative reading lives in search boxes instead. That is the more interesting fact of the two. The usage exists first and the dictionary follows.
A streak cannot express this arithmetic
Count ten three-day resolves as a streak and you get 3. Only the most recent survives; the previous nine are erased.
To say twenty you have to count cumulatively. An app whose engine is the streak switches off its own engine the moment it counts that way.
I wrote up why I left streaks out from a different angle, as a loss-aversion argument about the mechanic itself. This search data arrived at the same place from outside: the words people already use are cumulative.
The top-ranking posts all answer the first three
On the same day I skimmed the top result titles on Naver and Daum. Of the 16 lines that read like article titles, almost all pointed one way: five ways to overcome it, three ways to end it, the real reason you keep repeating it, why your new year resolution ends like this.
Every one of those answers the fix group or the cause group. Exactly one of the 16 speaks to the other side, the person who already knows this is not their first attempt.
Demand of 6 with a supply of 15 on one side. Demand of 5 with a supply of 1 on the other.
So I changed the copy
The deleted line treated the behaviour as a disease and addressed the three people looking for a cure, and those three already have 15 articles competing for them.
To the person typing “round N,” a promise to fix them is the tenth time they have heard it. What they need is not another promise but somewhere the previous nine rounds still exist.
Scope
This is one autocomplete pull from Google and Naver on 2026-08-16, one seed with eight suffixes, eight queries total, plus the top result titles read on the same day. The collection script lives in my repo.
I do not know absolute search volume. I do not know how many of the 36 are looking for an app, and I cannot promise autocomplete ordering is strictly frequency. Run it on another day and the list will shift somewhat. I read the top posts by title only, not in full.
I also do not know yet whether my app actually helps these people. Search data cannot answer that; only whether users come back can, and that number is not one I would boast about yet. When it is worth reporting, I will write it up.