Before you translate an app for a new market, the standard advice is: localize. Translate the UI, add local foods to the database, handle regional units. All of that sounds reasonable. I wanted to see whether it shows up where users complain without being asked.
So I pulled the most recent reviews for five calorie-tracking apps in the Bulgarian App Store — MyFitnessPal, Cal AI, Yazio, Lifesum, and Lose It! — 183 reviews in total, and read all 69 of the one- and two-star ones line by line.
Zero of them asked for Bulgarian. Zero mentioned a missing local food database. Zero mentioned translation quality.
That is the finding. It means less than it sounds like, and more than it sounds like, and the difference is the whole point of this post.
What I actually sampled
Apple publishes recent customer reviews as JSON per storefront, so this is easy to reproduce and hard to fake. One page per app, sorted by most recent, pulled on 13 November 2026.
| App | Reviews | 1–2 star | Window covered | Ratings on storefront |
|---|---|---|---|---|
| MyFitnessPal | 50 | 24 | Sep 2024 → Sep 2026 | 7,808 |
| Cal AI | 25 | 14 | Nov 2024 → Jun 2026 | 2,683 |
| Yazio | 33 | 11 | Apr 2016 → Oct 2026 | 1,828 |
| Lifesum | 45 | 10 | Jan 2015 → Oct 2026 | 972 |
| Lose It! | 30 | 10 | Mar 2017 → Nov 2026 | 618 |
Two things to note before the interesting part. The windows are wildly different — 24 months of MyFitnessPal and 141 months of Lifesum both fit on one page, so "the latest reviews" is not a shared time slice across apps (one page returned 25 for Cal AI and 50 for MyFitnessPal). I wrote about why that matters separately, because it quietly ruins a lot of comparisons built this way (what a sample of 200 actually covers).
And the storefront rating counts sit right next to the review counts to keep one trap visible: 7,808 people rated MyFitnessPal in this store, but only 50 left text. Ratings are a volume signal. Reviews are a reason signal. They are not the same population.
Where the anger actually goes
I sorted all 69 negative reviews into one bucket each, by primary complaint. Not by what the reviewer mentioned in passing — by what the sentence is about.
| Bucket | Count | Share of 69 |
|---|---|---|
| Money: paywall, charge, refund, price | 28 | 41% |
| Reliability: bugs, crashes, sync, login | 12 | 17% |
| Redesign: UI changed under existing users | 6 | 9% |
| Accuracy: wrong calories, wrong detection | 5 | 7% |
| Missing feature / region mechanism | 4 | 6% |
| Onboarding friction before value | 3 | 4% |
| Other (support, integrations, data loss, performance, vague) | 11 | 16% |
Money is not just the biggest bucket. It is bigger than the next two combined, and it splits into two completely different shapes depending on how old the product is.
New product, charge before proof. Six of Cal AI's fourteen negative reviews are about paying before being able to tell whether the thing works at all.
Не може да искате пари за нещо, което не можеш да пробваш и не знаеш дали работи.
(You can't ask for money for something you can't try and don't know whether it works.)
Alongside it: "You have to pay in order to use it", "I can't use it without paying, and I don't want to pay", and one review reporting a 35 lv charge that landed before the app would open the camera.
Old product, taking back what used to be free. MyFitnessPal and Lose It! produce the mirror image. Their negative reviews are not about being asked to pay. They are about things that were free two years ago and now are not — barcode scanning, macros, manual calorie entry.
It is horrible that they choose to move options that were a part of their free tier to the paid tier, you used to be able to scan barcodes and now I have to pay at least 50 euro a year.
Same bucket, opposite failure. The new app puts the bill in front of the trial. The old app moves the trial behind the bill retroactively, and the reviewer's complaint is not the price — it is the reclassification. When you analyze app pricing feedback, separating trial friction from feature-removal anger is the first cut that actually changes what you fix: a pricing page that explains tiers does not fix a reclassification, while a changelog that admits what moved does.
The single most useful pricing review in the set is a Russian-language one about Yazio, and it is about transparency rather than amount:
А самое смешное что если отказать от подписки то тебе сразу предлагают скидку в 75 процентов!!!
(And the funniest part: if you decline the subscription, they immediately offer you a 75% discount.)
Her inference is the one that costs you: "which means a lot of people were simply overpaying." Whatever the discount logic was meant to do, in this storefront it read as proof that the list price was fiction.
The redesign bucket is one product, three months
Six reviews land on "you changed the UI". Five of them are MyFitnessPal, all between July and September 2026, from users describing years of daily use:
The new Ui made the navigation more complex and difficult and is also so mundane like every other vibe coded app.
This is a different failure mode from a bug. Nothing is broken; the app just stopped being the one they knew. It does not show up in crash rates, it does not show up in retention for new users, and it only shows up in reviews from the people you most wanted to keep. I would not have found it by looking at a dashboard.
The closest thing to localization, and why it isn't
Two reviews in the entire set touch regional behavior. Both are Cal AI, both about measurement:
The option for logging food by measuring how many grams it weighs hasn't been working since the last update. Since then my experience with the app has felt horrendous as I'm european and measuring in Oz's is extremely inconvenient for me.
Read that again for what it asks for. Not Bulgarian. Grams. The user wrote in English, on a Bulgarian storefront, and the thing that broke their week was ounces. Storefront and language do not travel together — a distinction we keep strict when analyzing multilingual app reviews across storefronts — and here it exposes a units bug wearing a localization costume.
Barcode scanning shows up the same way — four complaints across four apps (scanner can't find the food, scan returns nothing an hour later, barcode moved behind Premium, barcode option gone after an update). Barcode lookup is the core interaction for packaged food in a European store, and it fails as a functional and commercial problem, not as a language problem.
What this can and cannot prove
Here is where I have to be careful, because "nobody complained about X" is the kind of finding that gets over-read in both directions.
What 69 reviews can support: a distribution. If 28 of 69 negative reviews are about money and none are about language, then money is the dominant complaint in this sample and language is not a visible one. That is a statement about what people chose to write, and it is enough to decide where to look first.
What they cannot support: absence. The reviews are the most recent ones a storefront returns, not a census. Zero localization complaints does not mean nobody wanted Bulgarian — it means nobody in this slice spent a review on it. I have made this mistake in the other direction before: treating a missing answer as an answer (what 200 reviews didn't say). Silence is evidence about salience, not about existence.
And they cannot support a trend. MyFitnessPal returned 50 reviews spanning 24 months. That is roughly two reviews a month in a store where 7,808 people have rated the app. You cannot build a month-over-month chart on two data points per month; you can absolutely build a "which bucket is biggest" ranking on 69. In a small storefront the useful question is distribution, not velocity.
The practical version
If you are about to enter a small non-English market, the first review pass should answer one question: which bucket is largest? Not "how is our sentiment", not "what is our rating trend". Those need volume you do not have yet.
And expect the answer to be boring in the same three places every time: what you charge for, what breaks, and what you changed. In this sample those three accounted for 67% of all negative reviews, and the market being Bulgarian changed none of them.
Localization is still worth doing. It is just not what people write about when they are angry — and if you budget for localization before you have read 69 angry reviews, you are guessing at which bucket is actually full.
Data: Apple's public customer reviews feed, Bulgarian storefront, one page per app sorted by most recent, collected 13 November 2026. Raw JSON for all five apps is available if you want to re-sort the buckets yourself — the classification is mine and reasonable people would move a few borderline reviews. I build ReviewFold, a tool for turning App Store and Google Play reviews into evidence-backed reports and side-by-side app comparisons, so my read on which buckets matter is not a neutral one.
Top comments (1)
The split between Cal AI's charge-before-proof and MyFitnessPal moving barcode scanning behind the paywall is a sharp one. Do you think the reclassification anger would also show up in the 3-star reviews, or only at 1-2 stars?