Why stats.fm and Last.fm miss your Spotify plays

We make an app in this category, so this comparison is not neutral and we are saying so before the first number. It is called Quilore, it is free, and it exists because of the first of the two complaints below. Read everything here knowing that.

What follows is not an opinion about which app is best. It is the review record, counted.

What we read

On 6 September 2026 we pulled the most recent Google Play reviews for five Spotify and music listening stats apps, in English, from the US store, sorted newest rather than by Play’s default relevance so the window is reproducible.

App	Installs	Store rating	Reviews	Window	Negative, with text
stats.fm	10,000,000+	4.00 of 111,929	400	27 May to 4 Sep 2026	199 (49.8%)
Last.fm	5,000,000+	2.27 of 90,956	400	26 Jun to 4 Sep 2026	357 (89.2%)
Receiptify	100,000+	4.19 of 16,421	400	26 Mar to 4 Sep 2026	68 (17.0%)
Trackify	100,000+	4.52 of 4,461	159	24 Apr 2024 to 13 Jul 2026	48 (30.2%)
must.fm	100,000+	4.11 of 994	80	5 Sep 2024 to 22 Aug 2026	34 (42.5%)

1,439 reviews, of which 706 are one or two stars carrying text. Only those 706 are counted below, because a rating with no words tells you nothing about why.

Two of the five could not produce 400 reviews at all. Trackify managed 159 across twenty-seven months and must.fm 80 across twenty-four. That is Play’s pagination limit meeting a thin review flow, and it is worth stating rather than hiding: a hundred thousand installs producing six reviews a month is a product people have stopped arriving at.

Every count below is a floor. The script that produced them requires both a subject word and a failure word in the same review, which undercounts on purpose. Where you see a number, the real figure is that or higher.

Complaint one: the record is incomplete

134 of 706 negative reviews, 19.0%. It is the largest theme in the corpus and it appears in four of the five apps.

App	Share of its negative reviews
Last.fm	90 of 357, 25.2%
stats.fm	34 of 199, 17.1%
Trackify	8 of 48, 16.7%
must.fm	2 of 34, 5.9%
Receiptify	0 of 68, 0.0%

The words people use are consistent across products that share no code. On stats.fm 2.9.0:

Used to be really reliable, now it misses huge chunks of songs you have streamed. Only counted 10 streams yesterday for an artist I listened to over 20 times.

On Last.fm 2.8, from a reviewer who moved to a third-party scrobbler instead:

App stopped scrobbling properly around June/July. Found myself double checking every listening session and having to manually scrobble at least half the songs.

And on Last.fm 2.6, the version most represented in that window:

Used to run reliably, however now just simply doesn’t register the songs I’m listening to. Currently missing a whole month of data ):

Why that complaint is structural

This is the part that makes the teardown worth writing rather than just unpleasant, and it is not a criticism of anyone’s engineering.

Every product above reads Spotify’s Web API. That API will tell you that a track played. It will not tell you for how long, so listening time has to be inferred from something other than measurement. And the endpoint that reports what you played recently is capped. Spotify’s own developer documentation says so:

This endpoint has only ever returned 50 items total.

So the app has to poll, and whatever happened between two polls has to fit in fifty items or it is gone permanently. A quiet Tuesday survives that. A long drive, a work shift or a party does not. Nothing in the app can recover those plays later, because the source no longer holds them.

One stats.fm reviewer worked out the answer without being told it:

All stats are incorrect until you request a new report from Spotify and upload it.

That is exactly right, and it is the whole reason a different architecture is possible. Spotify will send you your extended streaming history on request. That file carries the true millisecond duration of every play and reaches back to the day you joined, and reading it needs no API, no account and no polling. The cost is that it takes days to arrive, which is a real cost and not one to talk anybody out of.

Complaint two: paying to see your own listening

64 of 706, 9.1% overall, but the distribution is the interesting part.

App	Share of its negative reviews
Trackify	10 of 48, 20.8%
must.fm	7 of 34, 20.6%
stats.fm	40 of 199, 20.1%
Last.fm	7 of 357, 2.0%
Receiptify	0 of 68, 0.0%

Three products, three separate companies, three price models, and they land within 0.7 percentage points of one another. Roughly one negative review in five is about paying to see your own listening history.

the reason i downloaded the app was to see my listening minutes. didn’t know it was locked behind a paywall. i just wasted my time.

Charging for software is legitimate and we are not going to pretend otherwise. What the corpus actually shows is narrower and worse: the two complaints arrive together. People pay, and then find the first problem waiting for them.

you can only see your stats if you have the paid version and if you have the paid version the streams are completely random

What the star ratings do not tell you

Three things here would mislead anyone reading the store pages instead of the reviews.

Last.fm’s 2.27 is not a software rating. A charting-algorithm change drew an organised response from a fan community, and 47 of its 357 negative reviews (13.2%) are about that argument rather than about the app. A further 114 (31.9%) are under twenty-five characters and describe nothing at all. The missing scrobbles finding survives both, because it is counted only from reviews that describe a mechanism, but the headline number should not be quoted as evidence of anything.

stats.fm answers its reviewers. 374 of 400, 93.5%, including angry ones. We went in expecting to find neglect, because that is what the secondhand summaries of this category say, and the corpus does not support it. Whatever is wrong at stats.fm, nobody ignoring their users is wrong at stats.fm.

Receiptify is not really in this argument. 68 negative reviews, of which 57 are under twenty-five characters, and no hits at all on any theme in this teardown. It does one small thing, and the people who dislike it mostly do not say why.

What we built, and what we are not claiming

Quilore reads the export rather than the API, which removes the first complaint by construction rather than by being better at polling. It is free, with nothing behind a tier, which removes the second. Both claims are checkable rather than promised: it is at quilore.com, it runs in the browser, and the network tab will show you that nothing about your listening leaves the page.

Three honest limits on everything above.

This is one locale, us and English. A category with this many international users will read differently elsewhere, and we have not run those windows.

The windows are not equal. Two apps needed years to produce the reviews the other three produced in months, so their percentages describe a smaller and older sample.

And the counts are floors from a keyword method, not a careful reading of all 706 reviews. They are reproducible and they are conservative, which is the trade we chose. The raw corpus and the script that counted it both exist, so anyone disagreeing with a number can disagree with something specific.