Open research

What Steam's long tail actually does

Every chart here is computed from our own measurements of the full Steam catalog: 1.1 million monthly data points across 20,000 games with real commercial activity, going back to 2010. Recomputed nightly. Use it, quote it, link back.

A game stops decaying after year two

For every game we track, we compared the reviews it earned in each year of its life against the reviews it earned in its first year. The median tells a story that contradicts how most people talk about game revenue: the drop is brutal for two years, and then it stops.

That flattening is the whole thesis of this site in one number. A game that survives its third birthday keeps earning at roughly the same rate for as long as we can measure. Revenue does not fade to zero: it settles.

A 75% cut sells about the same as a 20% cut

We took every month of the last two years for games with tracked discount history, and compared each month against that same game's own months at full price. The lift is measured in review velocity, our proxy for units.

Read this one carefully. Deep discounts cluster around Steam's seasonal sales, where traffic is high for everyone, and games that discount heavily are usually older or slower to begin with. What the numbers do say is that going deeper does not buy proportionally more units, while it definitely costs more per unit sold.

The catalog keeps getting louder

Total reviews written per year across the games we track, and how many of those games were active in each year.

The 2020 jump is the pandemic, visible in raw review counts. The current year is partial.

How this is measured

• Review counts come from Steam's own monthly review histogram, which reaches back to each game's launch. We store every month rather than a rolling window.
• Discount history comes from IsThereAnyDeal, counting a month as discounted when a cut of that depth was live for at least five days.
• Every game is normalised against itself, so a big game and a small one weigh the same.
• Cohorts are 2012 to 2024 for the decay curve, so every cohort has room to age.
• Reviews are a proxy for units, not units. See the API docs for the full estimation chain and its error bars.