How often should I update content for AI search?

There's no universal number, but analyses point to a roughly three-month window where recency strongly helps citation eligibility. Prioritise refreshing your highest-value, most competitive pages on a regular cadence, and make the updates substantive (corrected facts, new data, real edits), not just a changed date.

Does freshness matter for every page?

No. It matters most for time-sensitive and competitive queries. Pricing, comparisons, 'best of' content, anything that changes. Stable, definitional content (like a glossary term) decays far more slowly. Spend your refresh budget where recency actually moves citations.

Content freshness and the 3-month citation cliff

In live AI search, age is a ranking signal. Engines that retrieve and cite pages at answer time lean toward content that's recently updated: and the corollary is harsher than most teams expect: citations decay as content goes stale.

The citation cliff

Analyses of AI citations keep surfacing the same pattern: a meaningful share of citations come from pages updated within roughly the last three months, and eligibility drops off after that. A page that earned citations on publish, then sat untouched, quietly loses them to fresher competitors covering the same ground.

The "publish a great post and move on" model. Fine for some evergreen SEO. Leaks citations in AI search.

The "publish a great post and move on" model. Fine for some evergreen SEO. Leaks citations in AI search.

Why engines favour fresh content

Live retrieval exists precisely to get current answers beyond a model's training cutoff. So when an engine assembles an answer, recency is a proxy for accuracy: a page updated last month is more likely to reflect today's reality than one from two years ago. For anything that changes. Prices, product lineups, "best X for Y," comparisons. Fresh wins.

How to refresh deliberately

Freshness isn't about churning everything constantly. It's about a cadence on the pages that matter:

Identify your high-value, competitive pages: the ones tied to buyer queries, and put them on a refresh schedule.
Make updates substantive. Correct facts, add new data or examples, revise stale claims. A bumped date with no real change is hollow (and engines increasingly look at actual content change, not just a timestamp).
Signal the update with an honest "last updated" date and genuinely revised content.
Leave stable content alone. Definitional and reference pages (a glossary term, say) don't need the same cadence. Don't spend refresh budget where recency barely helps.

Measure the decay

Because the cliff is gradual and invisible, you only catch it by watching citations over time. A page slipping out of answers months after publishing is the signal to refresh, and watching whether the refresh wins the citations back is how you know it worked. That ongoing read across engines is part of what measuring AI visibility (and Buffy Intel) is for.