Content freshness: definition and what the evidence shows

Content freshness describes how recent a page's substance is. Two different things get called freshness, and they are easy to confuse:

• Actual freshness: the content was written or meaningfully revised recently. Prices, specs, policies and dates in the text are current.
• Declared freshness: the page says it is recent. This lives in a visible "last updated" line, a schema dateModified, a sitemap lastmod or an HTTP Last-Modified header.

Every declared field is set by the publisher, and many CMSs set them automatically. So a recent date tells a crawler that something changed, not that anything worth knowing changed. For the full argument and evidence, read the guide on content freshness and AI citations.

Freshness touches AI answers at two stages.

Retrieval-on answers

When an engine searches the web, it may see a date for each result. Anthropic's web search tool documentation lists page_age, described as when the site was last updated, among the fields returned with each result. That confirms the signal reaches the model. It does not say the model prefers newer pages, and no vendor states that page age influences selection. The Claude visibility factors page grades this signal alongside the others.

Microsoft names freshness in its own guidance, and Bing has asserted that IndexNow helps AI systems reference the current version of a page. The IndexNow specification itself never mentions AI, and a successful ping confirms only receipt. See the Copilot citation factors for how that plays out on Bing-backed answers.

Retrieval-off answers

When a model answers from memory, freshness is capped by its knowledge cutoff. OpenAI's model docs list a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path. Anything you publish after a cutoff cannot appear in those answers, whatever its date says.

The case for keeping content current rests on accuracy, not on a ranking boost. If your pricing page still shows last year's price, an engine that fetches it can quote the wrong number. Stating the current fact plainly, on a page crawlers can fetch, gives the engine a correct sentence to use.

The experimental signal for dates is thin. The 252,000-trial factorial (arXiv:2605.25517) found that explicit price information and a recent timestamp helped consistently. That test used two candidate documents already placed in the model's context, with no live search. The 2026 critical survey of 45 GEO studies (arXiv:2607.14035) found no technique with a stable, cross-platform causal effect on discoverability.

The next move: update pages when the facts change, and leave the date alone when they don't.

• "Update your pages and AI will cite you." Seer Interactive reported in July 2026 that 75% of cited pages had been updated within a year. The study had no control group of uncited pages and no base rate for the wider web. If most pages carry a recent CMS date anyway, the figure carries no signal.
• "Bumping the date makes a page fresh." It changes a self-declared field. It does not change what the engine reads, and it makes your own freshness data meaningless. Our answer on changing the last-updated date covers this directly.
• "Schema dateModified is a freshness lever." It is one more declared field. Schema markup has no credible evidence as a citation lever.
• "Freshness has been disproved." It hasn't. It is untested at the stage that matters, so treat it as unresolved.

• Topical relevance: the factor with the most consistent support across studies, and a better use of editing time than date changes.
• Date consistency check: compare lastmod, headers, schema and visible dates so they agree with each other and with real edits.
• Content freshness and AI citations: the full guide to the evidence.

Does Google reward fresh content in AI Overviews? Google does not publish freshness as a requirement for AI Overviews or AI Mode. Its documentation says a page must be indexed and eligible to show with a snippet, and that no special optimization is needed. Whether recency affects which eligible pages get used is not stated. Keep facts current for accuracy, and do not expect a date change alone to move anything.

Is it risky to change dates without changing content? It doesn't produce a measurable benefit, and it degrades your own data. Once dates move without real edits, you can no longer tell which pages actually changed. It can also create mismatches between your visible date, schema and sitemap. Change the date when the substance changes, and keep every date field in agreement.

Which date fields do AI crawlers read? The common ones are the visible on-page date, schema dateModified, sitemap lastmod and the HTTP Last-Modified header. Anthropic's search tool returns a page_age value, but it does not document which field that comes from. All of these are declared by the publisher, which is why none of them proves a page is genuinely current.

How often should I update pages for AI search? Update when a fact changes: a price, a feature, a policy, a team detail. There is no evidence for a fixed refresh schedule as a citation lever. Prioritize pages that state facts AI engines are likely to quote, and check that those facts sit in the HTML crawlers can fetch without running JavaScript.