There is a field on every Steam Workshop item page that almost nobody talks about: the tag row. It sits under the title, it drives the "browse by tag" filters and the discovery sidebar, and it is the only piece of metadata the Workshop itself lets creators author apart from the title, the image and the description.
It is also, on 30% of the catalogue, empty.
Across 6,008,927 active Steam Workshop mods that carry at least one subscription — the entire live, in-use catalogue tracked by ModSignal, spanning 392 games — 1,806,103 mods (30.06%) carry zero content tags. Another 1,367,812 (22.76%) carry exactly one. The median active mod on the Steam Workshop is labelled with a single tag. Nine in ten carry five or fewer.
And the difference between being labelled and not being labelled is the single largest discontinuity in the dataset.
The ladder
Sort the catalogue by how many tags each mod carries and measure the share that reach 1,000 subscribers — the threshold the rest of this series uses for "this mod found an audience":
| Tags carried | Mods | % of catalogue | Reach 1,000 subs | % of all subs held | Favourites per sub |
|---|---|---|---|---|---|
| 0 | 1,806,103 | 30.06% | 2.05% | 3.99% | 0.034 |
| 1 | 1,367,812 | 22.76% | 12.59% | 18.46% | 0.044 |
| 2 | 970,741 | 16.15% | 17.38% | 17.54% | 2.005 |
| 3–4 | 1,009,954 | 16.81% | 23.30% | 28.15% | 0.050 |
| 5–7 | 543,454 | 9.04% | 34.76% | 23.77% | 0.090 |
| 8–11 | 187,956 | 3.13% | 21.53% | 6.61% | 0.055 |
| 12+ | 122,907 | 2.05% | 13.69% | 1.48% | 0.044 |
Two things stand out immediately.
The first is the cliff. Going from zero tags to one tag moves the success rate from 2.05% to 12.59% — a six-fold jump, achieved by adding a single label to a listing.
The second is that the ladder is not a ladder. It is an inverted U that peaks in the 5–7 band and falls away again. Mods carrying twelve or more tags do worse (13.69%) than mods carrying three or four (23.30%) — and there are 122,907 of them, so this is not a rounding artefact. The 9.04% of the catalogue sitting in the 5–7 band holds 23.77% of all subscriptions on the platform, while the 30.06% carrying nothing holds 3.99%.
A tidy story writes itself from that table: tag your mod, and tag it with five to seven labels. That story is wrong, and the rest of this article is about how we know.
The field is measured per game, not per mod
The first warning sign is not in the numbers but in the shape of the missingness. If "zero tags" meant "the creator declined to tag", it would be scattered thinly and randomly across every ecosystem. It is not. It is concentrated in whole games.
| Game | Active mods | Share with zero tags | Median tags |
|---|---|---|---|
| Opus Magnum | 6,617 | 100.0% | 0 |
| Alchemy Factory | 4,126 | 100.0% | 0 |
| Broken Arrow | 5,549 | 100.0% | 0 |
| PC Building Simulator | 6,470 | 100.0% | 0 |
| The Escapists 2 | 31,556 | 100.0% | 0 |
| Golf With Your Friends | 26,432 | 99.6% | 0 |
| Universe Sandbox | 71,818 | 79.6% | 0 |
| Human Fall Flat | 96,312 | 72.2% | 0 |
| Terraria | 101,298 | 69.2% | 0 |
| Space Engineers | 167,228 | 62.7% | 0 |
| People Playground | 79,227 | 62.5% | 0 |
Not a single one of The Escapists 2's 31,556 active mods carries a tag. Or Opus Magnum's 6,617. Or Broken Arrow's 5,549. No plausible creator-behaviour explanation produces a 0.0% tagging rate across five separate communities and a 100.0% rate in others. You can see the difference for yourself by comparing a fully-collected ecosystem like Garry's Mod against a completely uncollected one like The Escapists 2.
What is actually happening is that the tag field is collected per game. Steam publishes each game's tag vocabulary as its own list, and our crawler picks up those vocabularies one game at a time. In the games where we hold the vocabulary, tagging coverage approaches totality — Portal 2 sits at 57.3% untagged, Garry's Mod at 1.4%, while Terraria sits at 69.2% and Space Engineers at 62.7% on the wrong side of the collection line. In the games where we do not hold the vocabulary, every mod reads as untagged regardless of what its creator actually did on the Steam page.
This is the same class of trap this series has hit before with preview_url and description_short — see the thumbnail divide and the blank-page problem — and the same test exposes it. Cross-tabulate tagging against the two fields that are collected per mod:
| Tagged? | Preview? | Description? | Mods | Reach 1,000 subs |
|---|---|---|---|---|
| No | No | No | 1,531,424 | 0 of 1,531,424 |
| No | Yes | Yes | 274,679 | 13.45% |
| Yes | No | No | 597,943 | 0 of 597,943 |
| Yes | Yes | Yes | 3,604,881 | 22.81% |
The two "no preview and no description" rows contain 2,129,367 mods between them, and not one of them — tagged or untagged — ever reached 1,000 subscribers. That is not a finding about tags. Those are the rows our collector has not visited yet. The four cells are not four populations; two of them are an empty scaffold.
The honest comparison lives in the bottom two rows: holding both listing fields present, an untagged mod still reaches 1,000 subs 13.45% of the time against 22.81% for a tagged one. So there is a real gap inside the properly-collected population. It is just much smaller than the first table implied — and it is not the only thing going on.
The test that separates a lever from an artefact
The pooled gap inside the collected population is 22.81% versus 13.45%, a 1.70x difference. Every conventional control the earlier articles in this series used passes it: it holds in each publish year, it holds at every creator catalog size, and it holds inside individual games.
That is exactly the pattern that produced a false finding in this series before — the so-called publishing-hour illusion, and the crowding penalty that replaced it. Pooled comparisons of Steam Workshop mods pass every control you can think of and still lie, because creators are not interchangeable and neither are games. The only test that settles it is to compare the same creator against themselves, inside the same game: find creators who published both low-tag and high-tag mods into the same ecosystem, and ask which side did better.
We ran it, and it splits the story cleanly in two.
The zero-tag step survives. Holding creator and game fixed, comparing mods with no tags against mods with five to seven tags:
- 31,083 comparable creator-and-game pairs
- the tagged side wins on subscriber count in 29,795 of them (96.1%)
- geometric-mean subscriber ratio: 8.24x
The equivalent test on the first tag is even larger in sample: across 88,553 pairs comparing an untagged mod with a one-tag mod from the same creator in the same game, the one-tag side wins 79,561 to 8,122 — 90.7% of decided pairs — and the reach-1,000 rate moves from 2.69% to 7.16%, a 5.15x gap in magnitude.
The "tag generously" step does not survive. Run the identical test on the advice the inverted U implies — comparing mods carrying 1–2 tags against mods carrying 5–7 — and the effect evaporates:
| Test | Pairs | Low band (1–2 tags) | High band (5–7 tags) |
|---|---|---|---|
| All games, reach 1,000 subs | 31,351 | 29.04% | 30.03% |
| Well-collected games only | 12,237 | 30.20% | 29.50% |
Inside the properly-collected ecosystems the high-tag band is marginally behind. Across all games it is marginally ahead. Either way it is a wash on the threshold that matters, and it contradicts the clean 34.76%-versus-17.38% step in the pooled table by roughly twenty points. On subscriber magnitude the high band does keep winning — it is the larger side in 61.3% of the well-collected pairs, at a 1.49x ratio — but magnitude differences of that size are exactly what we would expect if well-established mods simply accumulate more labels over time than abandoned ones do.
So the inverted U is a composition artefact. It is not that five to seven tags cause success. It is that successful mods, and the mods of careful established creators, get labelled — and the two groups are not separable by tag count alone.
What the collected ecosystems actually show
Restrict the dataset to the 96 games where tag collection is near-total (fewer than 5% of mods untagged) and the picture is cleaner, and different:
| Tags carried | Mods | Reach 1,000 subs |
|---|---|---|
| 0 | 24,919 | 5.76% |
| 1 | 465,348 | 12.44% |
| 2 | 470,950 | 17.22% |
| 3–4 | 420,097 | 28.85% |
| 5–7 | 314,282 | 45.04% |
| 8–11 | 72,373 | 23.20% |
| 12+ | 27,301 | 13.01% |
The ladder is steeper and the 5–7 peak is higher, but the direction of the pooled evidence does not change: it still says 5–7 tags is the sweet spot. Only the paired test can tell you that the sweet spot is where well-run projects land, not where running a project well starts.
Even in these carefully-collected games, the paired result is a majority, not a law. Of the 38 ecosystems with at least 100 mods on each side of the comparison, the high-tag band wins in 28 of them — 74% — with an aggregate of 24.09% against 22.19%. A real edge, well short of the pooled table's promise, and one that comes with honest exceptions — the same discipline the short-title problem needed when its bracket-title effect inverted for Easy Red 2:
| Game | Reach 1,000 subs, 3–7 tags | Reach 1,000 subs, 1–2 tags | Median tags |
|---|---|---|---|
| MudRunner | 96.88% | 77.66% | 1 |
| House Flipper | 81.53% | 65.14% | 2 |
| Workers & Resources: Soviet Republic | 77.64% | 59.91% | 2 |
| Conan Exiles Enhanced | 70.99% | 27.07% | 1 |
| Ravenfield | 69.78% | 35.10% | 2 |
| Garry's Mod | 55.08% | 32.98% | 4 |
| SteamVR | 42.70% | 13.51% | 2 |
Those seven agree with the ladder. Not every game does — the sign test's ten losses are as real as its 28 wins, and the earlier articles in this series have published the counter-examples rather than the tidy version.
Where the missing labels are, and where they are going
The concentration of the missing tags is not only per game. It is also moving in one direction.
| Publish year | Active mods | Share with zero tags | Median tags |
|---|---|---|---|
| 2012 | 20,324 | 18.6% | 3 |
| 2015 | 149,760 | 24.3% | 2 |
| 2018 | 254,947 | 31.1% | 2 |
| 2020 | 428,459 | 25.2% | 2 |
| 2022 | 582,234 | 30.0% | 1 |
| 2024 | 795,982 | 33.4% | 1 |
| 2025 | 973,740 | 32.8% | 1 |
| 2026 | 836,638 | 31.9% | 1 |
The median active mod published in any given year from 2022 onward carries one tag. It was three in 2012. The share carrying none has drifted from the high teens to roughly a third. Two forces are at work and they are worth separating: the catalogue is growing faster than tag vocabularies are being collected, and the newer games arriving in the dataset are more likely to be games whose tag lists we do not yet hold. The 2026 figure of 31.9% is not a statement about how 2026's creators behave. It is mostly a statement about which games 2026's creators publish into.
The largest single vocabulary we hold belongs to Train Simulator Classic, whose 30,408 active mods carry a median of 10 tags each. It is the only major ecosystem in the 6-million-mod catalogue with a median above 8 — and it is a useful reminder that "how many tags is normal" is a property of the game, not of the workshop.
What this means if you publish mods
Three things, in descending order of how well the data supports them.
One. Put at least one tag on your mod. This is the finding that survives every control we have: same creator, same game, the tagged mod wins on subscribers in 96.1% of 31,083 paired comparisons and the untagged side is nearly never the winner. The step from nothing to something is the largest effect in this dataset, and it costs a few seconds on the upload form.
Two. Do not read the inverted U as an instruction to add tags until you hit five. That was our first conclusion and the paired test destroyed it: moving from 1–2 tags to 5–7 tags is a wash on reaching 1,000 subscribers inside properly-collected games. The band that peaks in the pooled table is where established mods live, not a destination you can walk to.
Three. If your mod appears in this dataset with zero tags and you know you tagged it on Steam, that is a bug in our collection, not a mistake on your part — and it is one we are working through game by game. The list of ecosystems with 100% untagged coverage is a to-do list, published above, and it is shorter than it was last month.
Method: 6,008,927 active Workshop items carrying at least one subscription, across 392 tracked games, from ModSignal's live Steam Workshop dataset (14.4 million item-tag joins against 20.4 million item-tag rows). "Tag" means a workshop_item_tag join to a workshop_tags row for the item's own game; Steam maintains a separate vocabulary per game, so counts are not comparable across ecosystems and are never merged. "Well-collected games" are the 96 ecosystems with at least 3,000 active mods and fewer than 5% untagged. Paired tests group by (creator_steam_id, game_id) and keep only creators and games with at least one mod on each side of the comparison; "wins" compares mean log(subscribers + 1) per side, with ties excluded from the reported share. All figures are a live snapshot and move as the Workshop graph refreshes — current as of September 26, 2026.