Steam’s appreviews endpoint (num_per_page=0 for a summary total, filter=recent for a paged
sample) returns one API and five different questions depending on how you slice it. This page is
what we have learned running that endpoint across nine games: a small sample can invert a verdict,
a purchase route can move a score either direction, and two boolean flags that sound useful turn out
to be nearly empty.
In August this site said Project Zomboid reads 40% positive in Simplified Chinese. That was true of the 100 most recent Chinese reviews at the time. Across all 29,030 of them, queried again on 2026-09-13, the figure is 86.1%.
Trap one: a 100-review sample can read backwards
| Game | Language | Recent 100 (27 Aug) | All-time (13 Sep) | n | Gap |
|---|---|---|---|---|---|
| Project Zomboid | French | 97% | 95.4% | 11,104 | -1.6 |
| Project Zomboid | Chinese (Simpl.) | 40% | 86.1% | 29,030 | +46.1 |
| Escape from Tarkov | Portuguese (BR) | 79% | 58.0% | 1,320 | -21.0 |
| Escape from Tarkov | Chinese (Simpl.) | 31% | 23.3% | 15,720 | -7.7 |
| Mistfall Hunter | Portuguese (BR) | 59% | 69.7% | 756 | +10.7 |
| Mistfall Hunter | Russian | 17% | 40.3% | 2,322 | +23.3 |
| Corsair Cove | Chinese (Simpl.) | 69% | 77.3% | 578 | +8.3 |
| X4: Foundations | Portuguese (BR) | 89% | 90.4% | 332 | +1.4 |
| X4: Foundations | Chinese (Simpl.) | 71% | 77.9% | 3,094 | +6.9 |
| Songs of Syx | French | 99% | 97.6% | 167 | -1.4 |
| Songs of Syx | English | 85% | 94.1% | 6,692 | +9.1 |
A most-recent-100 sample answers what do people writing now think. An all-time total answers what has everyone ever thought. For a game five weeks old those are nearly the same question. For a game twelve years old they can be 46 points apart.
The Project Zomboid case is the clearest. The August sample landed while Chinese-language reviews were dominated by Build 42 mod breakage — the topic pass found half of the negatives were about updates and half about mods. The all-time count says that window was an exception: 86.1% of 29,030 Chinese reviews are positive, seven points under English rather than forty-four.
The gap runs both ways. Tarkov’s Simplified Chinese sample read 31%; the all-time figure is 23.3%, lower — the recent Chinese reviews of Tarkov were kinder than the history. Mistfall Hunter’s Russian floor, 17% in August, is 40.3% across all 2,322 — still the lowest language for that game, but not by the margin the sample suggested.
Trap two: purchase route moves the score, and not in one direction
There is a folk theory that key-activated copies (retailer, bundle, giveaway, press or developer key) review worse than store purchases — less invested buyers, cheap bundles. Queried all-time on 2026-09-13, the theory is right for four games here and wrong for three. Counter-Strike 2 is excluded: it has been free to play since 2018 and the split does not mean the same thing there.
| Game | Store positive | Key positive | Key minus store |
|---|---|---|---|
| Mistfall Hunter | 53.9% | 45.9% | -8.0 |
| Escape from Tarkov | 49.9% | 46.0% | -3.9 |
| Songs of Syx | 94.5% | 89.4% | -5.1 |
| Project Zomboid | 94.3% | 92.2% | -2.1 |
| Corsair Cove | 85.6% | 85.2% | -0.4 |
| The Blood of Dawnwalker | 85.2% | 87.0% | +1.8 |
| X4: Foundations | 79.4% | 82.5% | +3.1 |
Mistfall Hunter, Tarkov and Songs of Syx are games whose reviews turn sour with hours played; their key buyers read harsher. X4 and (now more mildly than at launch) Dawnwalker are games where the early hours are the hard part; their key buyers read kinder. Corsair Cove is the control: no gap at all. Whether key buyers are a genuinely different population or simply play a different number of hours before writing, the summary endpoint cannot say — and Dawnwalker’s gap nearly closing in nine days (from +5.6 on 2026-09-05 to +1.8 here) shows how fast that number can move on a young game.
Trap three: the biggest review bases write the least
Median review length, share under 50 characters and share past 500, from the most recent reviews of
eight games pulled 2026-09-05 with filter=recent (up to 8,000 per game; not re-pulled for this
page — see limitations):
| Game | Median length | Under 50 chars | 500 chars or more |
|---|---|---|---|
| Counter-Strike 2 | 14 | 82% | 1.2% |
| Escape from Tarkov | 35 | 58% | 6.4% |
| Project Zomboid | 37 | 57% | 5.8% |
| Mistfall Hunter | 84 | 38% | 14.6% |
| Songs of Syx | 92 | 36% | 16.0% |
| Corsair Cove | 126 | 32% | 21.0% |
| X4: Foundations | 128 | 30% | 22.1% |
| The Blood of Dawnwalker | 168 | 27% | 23.3% |
Counter-Strike 2, Tarkov and Project Zomboid — the three largest review bases on this site — have medians of 14, 35 and 37 characters and are more than half one-liners: a vote, not an argument. The other five have medians between 84 and 168, and a fifth to a quarter run past 500 characters. A topic pass has far less to read on the vote games; the rating means something closer to a poll on CS2 and closer to a summary of arguments on X4.
Trap four: two review flags that are almost always empty
Steam stores primarily_steam_deck and received_for_free on every review. Across the same
2026-09-05 pull of eight games (45,357 reviews for the Deck flag; the same set for the free flag),
both are nearly unset: 93 Deck reviews (0.21%, highest X4 at 0.61%) and 354 free-copy reviews
(0.8%, highest Tarkov at 1.6%, reading 16 points kinder than the rest of its set on that small base).
Neither filter would move a rating by a tenth of a point on these eight games; they are documentation
of what is not there, not a lever worth pulling.
Trap five: whether the reviewer is still playing
The field Steam records but the store page never shows: whether the reviewer had any playtime in the two weeks before writing. Refreshed today for the two games this site tracks most closely — X4 and Counter-Strike 2 read the same direction the 2026-09-05 pull found for five games:
| Game | Still playing | Stopped | Gap | Read |
|---|---|---|---|---|
| Escape from Tarkov | 57.2% | 40.3% | +16.8 | 2026-09-05 |
| X4: Foundations | 86.9% | 72.2% | +14.7 | 2026-09-13 |
| Corsair Cove | 92.0% | 82.9% | +9.1 | 2026-09-05 |
| Mistfall Hunter | 50.1% | 41.9% | +8.2 | 2026-09-05 |
| Songs of Syx | 92.1% | 87.3% | +4.8 | 2026-09-05 |
Every game reads higher from reviewers who were still playing when they wrote. What “stopped” means depends on the game: Corsair Cove’s stopped reviewers are finished, not unhappy, at 82.9%; Tarkov’s stopped reviewers, for a game with no ending, are closer to a complaint at 40.3%. The same cut, applied to a story game and a live one, measures two different things.
What this changes on this site
Language tables here use all-time summary totals where the question is “how does language X rate this game”, and recent samples only where the question is explicitly about now — the per-game all-time tables started with X4, Songs of Syx and Corsair Cove. Purchase route, review length and recent-playtime status get their own cut wherever a per-game piece has the review-level data to support it, rather than being folded into one aggregate score.
What wasn’t re-queried for this page
Review length and the two review flags are the original 2026-09-05 pull, not re-queried for this
page — a fresh 8,000-review pull per game was out of scope here, and the finding (both flags are
under 1%) is not the kind that moves week to week. Steam’s recent filter is by creation time and
single-page pulls are taken as reported.
Sources
- Steamworks documentation, “User Reviews – Get List” — the public
appreviewsendpoint and itslanguage,purchase_type,num_per_pageandfilterparameters - Example all-time query, Project Zomboid in Simplified Chinese — re-run 2026-09-13 for all eleven sample-vs-total cells and all seven purchase-route rows
- Example page, most recent Counter-Strike 2 reviews — the 2026-09-05 review-level pull behind the length and flag tables
