How Kitted.gg tracks Warzone loadouts: what is watched, what counts as a confirmed build, how the one statistic is computed, and what the known error modes are. Everything on the site is produced by this pipeline; nothing is written by hand.
Five sources feed the pipeline. Twitch clip vision: new clips from each tracked streamer are scanned and any gunsmith screen is read by computer vision. Live scan: when Twitch reports a streamer has gone live, the stream is sampled immediately and again a few hours later. YouTube: recent videos are scanned the same way, and video descriptions are parsed for loadout text. Chat commands: loadout commands in the streamer's own chat are read, subject to the demand-consensus rule below. CODMunity pins: streamers who publish their builds there are read directly.
A build is confirmed only when both slots are known and each slot has at least four attachments read from the gunsmith. For chat commands, a reading needs consensus: one asker can enrich an existing build, two independent askers are needed to introduce a new weapon, and three are needed to overturn a build that was read from the gunsmith. A kill-feed or HUD read with no attachments never overrides a built loadout.
Every slot carries the timestamp of its newest evidence. A streamer counts toward the public statistics only if they have a complete build confirmed in the last 14 days. Streamers who are standing by (no complete build inside that window) still have a page, but they are excluded from every count and from the consensus meta, and their page is marked not-for-indexing once the last confirmed build is older than 45 days.
Pros means the Warzone streamers Kitted tracks who have a complete build confirmed in the last 14 days. A streamer counts once per weapon, whichever slot they run it in. Pick rate is that count divided by that denominator. The same number appears in page titles, first sentences, tables, JSON-LD and llms.txt; the canonical table is at kitted.gg/meta/most-used-weapons.
Weeks run Monday to Monday and only completed weeks are published. Each week is reconstructed from the raw evidence with the same rules as the live page: the newest complete build per streamer before the week ended, inside the same 14-day floor. Trends report share rather than raw counts because the roster grew over the window, so a weapon can post rising counts while losing ground.
OCR truncation: a gunsmith read can cut an attachment name short; truncated reads are merged with the full name by variant matching. Misreads: a weapon read that does not overlap the streamer's current loadout is dropped rather than published. Build codes: a code is attributed to the one named streamer it was read from and is never averaged across streamers. Every build links to the clip or video it was read from so a reader can check it.
Kitted does not track settings, sensitivity, field of view, hardware, or player statistics. It does not rank weapons by damage, time to kill, or any opinion. The site reports what the tracked pros run, and only that.
Cite as: Kitted.gg pro pick rate, with the date shown on the page and the page URL. Example: "Kitted.gg pro pick rate, September 8, 2026, kitted.gg/meta/most-used-weapons". The CSV at kitted.gg/meta/most-used-weapons.csv carries the same figures with an as_of timestamp.
About Kitted.gg, the current Warzone meta, weekly adoption trends.