A query freshness window is the recency limit an AI search tool attaches to a single background search: how recently a page must have been updated to be preferred as a source for that specific sub-query. It is chosen by the topic, not by the site — the same page can be fresh enough for one query and stale for another.
The clearest example is ChatGPT's 2026 pipe-delimited search format, where each search line carries a freshness value in days as its third field. Observed values scale with how fast the facts change: search consultant Suganthan Mohanadasan reported roughly 2 days for stock prices, 7 for sports results, 30 for commercial products, 90 for earnings data, and 365–3,650 days for evergreen forum content (a directional, single-account observation from August 2026 — treat the exact numbers as a snapshot).
Freshness windows matter because they turn "keep content fresh" from a vague instruction into a per-query threshold. A page outside its window can still be fetched — sometimes as a silent source that shapes the answer without a link — but it is not preferred for the citation. It is the retrieval-time cousin of content decay: the window is the engine's recency rule at search time, and decay is what you observe downstream if you never refresh. The practical response is to match your refresh cadence to each page's window — monthly for commercial pages, rarely for definitions.