The Steam Workshop's Off-Platform Link (2026): Only 6.9% of Mods Put a URL in a Description — and 4 in 10 of Those Point Only Back at Steam

Every listing on the Steam Workshop is a box. There is a title, an image, a set of tags, and one free-text field for a description. This series has now measured the title (its length, its uniqueness, what script it is written in), the image (present or not), the tags (how many, which ones) and the description text (present or not, unique or pasted). What none of them asked is the one thing a creator is most inclined to do with the description box: put a link in it.

So we scanned every description we hold and looked for http. Then we looked at what the link pointed to, because that turns out to be where the story is.

The headline, and why it is smaller than it looks

Of 3,874,980 active Workshop mods that carry at least one subscription and a non-null description, 267,966 (6.92%) contain a URL of any kind.

That is the number. Now split it by destination:

153,914 mods carry a link that genuinely leaves Steam, with no Steam link at all — 3.97% of the description-bearing catalogue. Add the 2,516 that carry both, and the off-platform population is 156,430 mods, or 4.04%.

The reason to bother with that split is that the first number is not the finding. A steamcommunity.com URL in a Workshop description is almost never a creator reaching its audience. It is a creator storing something — a preview image hosted on Steam's own CDN, a link back to their own workshop page, or a cross-link to a friend's or a required mod's page. Those are not distribution links and they tell you almost nothing about a creator's relationship with an outside audience. The genuinely interesting population is the 156,430 that point somewhere Steam does not own.

The channels they choose:

Read the last column carefully, because it is the trap this series has now walked into five times.

The pooled gap, and the paired test that survives it

Pooled, the difference is enormous. Mods with any URL reach 1,000 subscribers 39.44% of the time against 20.83% for mods with none — a 1.89x gap. On the honest off-platform split the pooled gap is even bigger: 47.53% vs 28.05%. Support links look like the strongest lever in the dataset at 61.11%.

Then you hold the creator and the game fixed, which is the only control that has ever separated a real lever from a composition artifact in this catalogue. Across 9,089 creator-and-game pairs where the same person published both a link-less mod and an off-platform-linking mod in the same ecosystem:

+15.1 points, a 1.44x advantage — and this is the first time in this series that the paired test has come out decisively positive. The publishing hour, the crowding penalty and the copy-paste descriptions all collapsed or reversed under this control. This one holds, and it holds in the same direction as the pooled number.

It holds through the length control too. Restricting both sides to descriptions of 100–162 characters — the width band where a link is not simply a proxy for "wrote more text" — leaves 1,675 pairs:

A 1.52x gap, wider than the unmatched version. And the per-pair sign test agrees: of 1,675 pairs, the off-platform side wins 389 and loses 347, with 939 ties — a 52.9/47.1 split among decided pairs. That is a modest majority, not a landslide, and it is the honest number to quote: the paired evidence supports a real but moderate association, not the 1.89x the pooled table suggests.

What this is actually measuring

Here is the caveat, and it is not small. "Description contains a URL" is correlated with three things that all cause success on their own: the creator ran the mod long enough to build a community, they cared enough to write a real description, and they are operating as a semi-professional outfit rather than a one-off uploader. A link is a marker of a creator who is treating the Workshop as one channel among several. This dataset cannot separate "linking out helps a mod" from "creators who link out are the ones who were going to succeed anyway."

What it can say is that the association is the wrong shape to be pure noise. The magnitude rose under length-matching, the pooled and paired numbers point the same way, and the per-channel ordering is coherent — the links that indicate an established operation (payment, source hosting, non-Western community platforms) clear 1,000 subscribers at 43–61%, while a YouTube link sits at 32.9%, barely above the 28.05% baseline for a Steam-only listing.

The cleanest way to read the channel table is as a map of what kind of creator writes what. A Discord invite is the most common off-platform link on the platform (15,675 mods) and the least discriminating — it is what you write when your mod has users. A Patreon or Ko-fi link is rare (5,107 mods) and sits at 61.11%, because it exists only for creators whose audience is already large enough that someone might pay. That is a statement about who has a tip jar, not about what a tip jar does.

The geography of linking out

Off-platform-link rates are not evenly distributed, and the map is legible:

Dungeon Alchemist and Rust effectively require an outside link, because both are ecosystems where the Workshop listing is a pointer rather than the product — a Dungeon Alchemist map is a file you get from the creator's own site, and Rust's Workshop lane is full of unlock-abuse and gift uploads whose descriptions are a redirect. Wallpaper Engine is the outlier this series meets in every article: nothing about it behaves like the rest of the Workshop.

The other end is where a link is pointless because the mod is self-contained. Left 4 Dead 2 is the only major ecosystem where linking out is neutral to negative in the per-game split — 32.15% for its 28,445 linking mods against 32.75% for the 142,602 that do not. It is the largest single population of linking mods on the platform and it is the one place where a link predicts nothing. Project Zomboid is nearly as flat (32.80% vs 29.89%). Both are games where the Workshop is the distribution channel and a good mod needs no forwarding address.

The split holds in 24 of the 30 largest ecosystems in the direction the pooled data predicts. The exceptions — Left 4 Dead 2, Easy Red 2 and Empyrion among them — are published above rather than quietly dropped, because a 30-game sweep that agreed unanimously would be a sign the control was doing nothing.

What it means

For a creator: the link is not the lever, but the practice is not neutral either. The paired evidence says a mod that links out clears 1,000 subscribers about 15 points more often than the same creator's mod that does not, and about 21 points more when you hold description length fixed. Read that as the signature of a creator who has somewhere to send people, not as a directive to paste a Discord invite into an old listing. Adding a URL to a mod no one has installed changes nothing.

For anyone reading a Workshop statistic: this is the fifth field where the pooled average overstated the case. The pooled table says 1.89x. The paired test says 1.44x. The length-matched paired test says 1.52x. Two of those three are honest; the first one is the one that would get quoted. The rule this series keeps re-learning is that any per-listing attribute a prolific creator can assign in bulk will carry a pooled advantage that is partly a productivity effect — and the way to find out how much is to compare a creator against themselves.

The operationally interesting finding is the self-reference. Steam's own CDN and workshop URLs account for 42.6% of every URL on the platform, and 111,536 listings — 41.6% of all URL-carrying mods — link to Steam and nowhere else. The Workshop's description field is, for the single largest block of users that put anything in it, being used as a place to store an image or point back at Steam. The mods that treat it as a way to reach an outside audience are about four in a hundred.

Related reading: the blank-page problem on the 38.6% of listings that write nothing at all, the copy-paste descriptions on the 189,543 mods sharing a blurb, and the visitor-counter gap on the 80 ecosystems where the reported subscriber count exceeds the visitor count.

Method: 3,874,980 Steam Workshop items tracked by ModSignal as of 1 October 2026 — every item that is is_active, carries at least one subscription, and has a non-null description. A "link" means the description matches https?://; a "Steam link" matches steamcommunity.com or steampowered.com; "off-platform" means a URL is present and no Steam URL is. Channel regexes are keyword matches (Discord, Patreon/Ko-fi/PayPal/Boosty/Gumroad, YouTube, GitHub/GitLab, Nexus/ModDB/CurseForge, VK/Telegram/QQ/Bilibili) and overlap, so the channel column does not sum to the off-platform total — Discord is checked first in the ordering discussion and the largest single channel is Discord at 15,675. "Reached 1,000 subscribers" means subscriptions_current >= 1000 at a single consistent read; it is a point-in-time threshold, not a tracked trajectory. The paired tests group by (creator_steam_id, game_id) and keep only creators who published at least one mod on each side; the length-matched version additionally restricts both sides to descriptions of 100–162 characters. Paired results are unweighted aggregates over the qualifying mods, not per-creator medians, and they under-cover mass-publishers whose catalogue is almost entirely one-sided — which is a reason to treat the paired number as the conservative one. Descriptions in this dataset are truncated at the point of collection: ModSignal stores Str::limit($description, 160), so descriptions are capped at 163 characters including the ellipsis — which is why 1,183,432 mods sit at exactly that length. A URL placed later in a long description is therefore never observed, so every link rate here is a lower bound, and the loss is concentrated in the verbose, heavily-formatted listings most likely to carry several links. Counters are the last reported values and are never decremented by Steam. Queries were run against the live Postgres catalogue; a re-run will differ by a handful of items.