ChatGPT now tells its search tool how recent a page must be to count. In its 2026 pipe-delimited query format, the third field of every background search is a freshness window in days — roughly 2 days for stock prices, 30 days for commercial products, and 365 to 3,650 days for evergreen forum content. A page updated outside the window for its query type can still be fetched, but it is not preferred. This is a step-by-step guide to which windows apply and how to keep your pages inside them.
Last reviewed: 23 August 2026. The window values below were observed by search consultant Suganthan Mohanadasan (suganthan.com) across a small set of questions on a single account in August 2026. Treat the existence of a per-query recency field as the durable finding and the exact day counts as a directional, single-observer snapshot — attribute "Suganthan, August 2026" and date any figure you reuse. This assumes you already accept that freshness affects citations; the question here is how recent, for which pages.
What is a freshness window in ChatGPT search?
It is the recency ceiling ChatGPT hands to each sub-search. When the model fans a question out into many searches, each line carries a number of days in its third field — a query freshness window. A window of 30 means "prefer pages touched in the last 30 days for this sub-query." Short windows enforce recency hard; long windows effectively switch the recency filter off. The window is chosen by the topic, not by you, which is why the same site can be perfectly fresh for one query and stale for another.
What freshness windows does ChatGPT use by query type?
They scale with how fast the underlying facts change. The observed values, from most to least time-sensitive:
| Query type | Observed freshness window | What it implies |
|---|---|---|
| Stock prices | ~2 days | Only near-live data qualifies |
| Sports results | ~7 days | This week's outcomes only |
| Commercial products / pricing | ~30 days | Refresh monthly to stay eligible |
| Earnings / financial guidance | ~90 days | Quarterly cadence is enough |
| Reddit / evergreen forum content | ~365–3,650 days | Recency barely filters; age is fine |
Source: Suganthan, August 2026 (small single-account sample; directional). The pattern is intuitive: the faster a topic's facts move, the tighter the window. The actionable row is commercial products at ~30 days — most brand-relevant pages (pricing, plans, comparisons, "best for X") sit in that bucket, which means a page that has not been genuinely updated in a quarter risks falling outside the window exactly where buying decisions happen.
How often should you refresh each page type?
Map your pages to their window and set the cadence to match — no more, no less:
- Pricing, plans, and comparison pages → monthly. These hit the ~30-day commercial window. A substantive monthly review (verify prices, update the comparison, note what changed) keeps them eligible.
- News, launches, and time-sensitive pages → as events happen. Short windows mean stale is worse than absent; update on the news, not on a calendar.
- Financial or quarterly-data pages → quarterly. The ~90-day earnings window means a quarterly refresh is sufficient; monthly churn adds no benefit.
- Evergreen explainers and glossary/definition pages → rarely. Long windows plus slow real-world change mean these decay slowly. Refresh when the facts actually change, not on a timer — over-churning stable content wastes your budget and can erase accrued authority.
The rule of thumb: spend your refresh budget where the window is short and the query is commercial. A page in a 30-day window fighting for a buying query earns far more from a real monthly update than a definition page in a 10-year window ever will.
Does this contradict the three-month citation cliff?
No — they are two different stages of the same story, and it is worth being precise so the corpus does not appear to disagree with itself. The freshness window is a retrieval filter applied by one engine (ChatGPT) at the moment it searches, expressed in days per query type. The three-month citation cliff is an observed decay pattern — across engines, a page's citations tend to fade after roughly a quarter without updates.
The freshness window is the engine's recency rule at search time; the citation cliff is the decay you observe downstream if you never refresh. They are cause and effect, not a contradiction.
Both point to the same discipline: keep competitive, time-sensitive pages genuinely current. The window explains why a specific ChatGPT search skips your stale page today; the cliff explains what the cumulative neglect costs over a quarter. Neither implies you should churn evergreen content that decays slowly.
How to check your pages against ChatGPT's windows
A quick audit you can do without special tools:
- Classify each key page by query type. Is it commercial (30-day), time-sensitive (days), quarterly (90-day), or evergreen (long)? That sets its target cadence.
- Check the visible "updated" date and the real content behind it. A date bump with no substantive change does not count; ChatGPT rewards updated content, not updated timestamps.
- Prioritise the short-window commercial pages. These are where falling outside the window costs citations on buying queries — fix these first.
- Measure citations before and after a refresh. Because of the known crawl-to-index-to-cite lag, watch the trend over several weeks, not overnight.
The honest takeaway: ChatGPT's search tool now encodes recency as a hard, per-topic parameter, so "publish and forget" fails fastest exactly on your commercial pages. Knowing which of your pages sit in short windows — and whether your refreshes actually move citations — is the loop Buffy Intel is built to close. Questions: [email protected].