Freshness wins: content from the last 13 weeks captures half of all AI citations

Content published or materially updated in the last 13 weeks earns roughly half of all AI citations, according to 2026 cross-engine citation analyses. That single number changes what freshness is: not a property of a page you tick off, but a property of your whole library — how many pages your site can keep inside a rolling three-month window at the rate you actually publish.

The part that trips most WooCommerce teams isn't writing, it's arithmetic. Every priority inside Contexta is ranked from a site's own Search Console data, and the freshness question we keep hitting there is never "is this page fresh" — it's "how many of these pages can possibly be fresh at once, given the cadence this team can sustain." What follows is that math, the decay curve underneath it, and the honest limit on how far cadence can carry a site.

Why does content from the last 13 weeks capture half of AI citations?

Because AI engines retrieve live at query time rather than serving a ranked index, and recency is the cheapest available proxy for "still true." Across 2026 analyses of AI citation sets, content under 13 weeks old accounts for about half of all citations, and content under 30 days has been measured earning citations at roughly 3.2× the rate of older pages.

The mechanism is worth being precise about, because it is not a ranking boost. When an assistant fans a question out into retrieval queries, it assembles a candidate set and then has to choose which passages to quote. Publication and update recency acts as a filter on that candidate set — a page that looks stale gets dropped before quality is ever assessed. This is a different scoring layer from the one deciding which source pool each engine draws on in the first place; you have to survive both.

What does the citation decay curve actually look like?

It rises fast and falls faster: most pages hit peak citation volume somewhere in the 30–90 day range after publication, then decline steadily, with near-total dormancy by roughly the one-year mark unless the page is substantively updated. The curve is not a slow fade — the steepest part of the drop typically lands in months three through six, which is exactly when most teams have stopped thinking about the piece.

Half-lives differ sharply by engine, which matters if one assistant sends you most of your AI traffic. From 2026 citation-half-life measurements:

The apparent contradiction in that last row is real and worth flagging: Perplexity keeps individual pages alive longer, yet it is also the engine most biased toward brand-new content overall. Both are true because it retrieves the live web on every query. We unpack why the engines diverge like this in why ChatGPT and Perplexity cite different sources.

How is AI citation freshness different from on-page SEO freshness?

Classic SEO freshness was a per-page tiebreaker on query-deserves-freshness topics; AI citation freshness is a portfolio-level eligibility filter that applies to nearly every query type. The old signal asked whether this page deserved a recency bump. The new one asks whether any of your pages are recent enough to enter the candidate set at all.

The practical difference shows up in how the two respond to a date change. Updating dateModified while leaving the body untouched has historically moved the classic freshness signal a little. It does not survive AI retrieval, because the engine re-fetches the page and compares substance — one 2026 study tracking a 90-day window found pages receiving genuine content, data and structural updates monthly produced around 4× the citation events of pages given cosmetic date changes only. We have watched the same thing on client sites: a bulk "updated" sweep across a category produced no measurable movement in AI referrals, while three genuinely rewritten pages did.

The other difference is who the clock belongs to. In classic SEO, freshness attaches to the page. In AI retrieval, it effectively attaches to the query — a volatile question ("ChatGPT checkout fees 2026") applies a hard recency filter, while a stable one ("how to measure jeans inseam") barely applies one at all. Your cadence should follow query volatility, not the calendar.

How many pages do you need inside the 13-week window?

Multiply your monthly publish-or-refresh cadence by three — that product is the total number of pages your site can hold inside the window at any moment, and nothing else changes it. A store touching five pages a month keeps 15 fresh pages, permanently. Not 15 this quarter and 30 next; 15, forever, because pages leave the window at exactly the rate you add them.

of the library sits inside the 13-week window — the other 92% competes on authority alone

That is the number most freshness advice never states, and it reframes the whole problem. Three consequences follow, and none of them are about writing better:

• Fresh inventory is capped by throughput, not effort. Working harder on the same five pages doesn't add a sixth to the window.
• Site size is a liability here. A 40-page site at five a month keeps 38% of itself fresh; a 400-page catalog keeps 4%. The bigger library needs a far higher cadence for the same eligibility rate — or has to concede that most of it will never be freshness-eligible.
• Cadence beats volume. Twelve pages published in one January burst and then abandoned leave you with an empty window by May. The same twelve spread across the year keep three or four live at all times.

Which pages should get your refresh cadence?

The ones losing the most clicks on volatile, high-impression queries — which is a Search Console question, not an editorial one. Because the window is capped, choosing which pages occupy those slots is the highest-leverage decision in the whole strategy, and picking them by gut reliably wastes them on pages nobody was going to cite anyway.

The signals we look for, in order: pages with high impressions and collapsing CTR (the query moved, your answer didn't); pages ranking 11–20 on queries that carry a year or a version number; and pages whose queries have shifted wording over the last two quarters. Those three groups are where a substantive rewrite converts into citations, because the query itself is volatile enough for the recency filter to be doing real work.

This is the job Contexta's Problem Map does — it imports a site's Search Console data and ranks pages by estimated lost clicks per month, so the refresh queue is ordered by measurable loss rather than by which post someone remembers being proud of. It won't tell you a query is volatile; it will tell you which pages are bleeding, which is where you start. The underlying method is the same one behind reading your own Search Console data properly, and it's worth doing by hand once before automating it.

When is chasing freshness the wrong move?

When your pages aren't retrievable in the first place, or when your queries are genuinely stable — in both cases cadence spends real effort for close to zero citation gain. Freshness is a filter applied to candidates; if GPTBot or PerplexityBot can't fetch the page, or the content only exists after JavaScript runs, there is no candidate to filter and the refresh changes nothing.

Two honest limits on the whole argument. First, Google AI Overviews shows the weakest freshness bias of the major surfaces — citation patterns there track traditional organic age profiles more closely, so an authoritative older page can keep earning AI Overview citations without a refresh. If Google surfaces are your main AI channel, cadence is a secondary lever behind authority. Second, evergreen commercial pages with stable queries decay far more slowly than news-adjacent content; forcing a quarterly rewrite onto a product category page that hasn't changed is churn, and churn has its own cost in review time and internal-link breakage. The pages worth the slot are the ones where the world actually moved — the rest are better served by fixing the structural reasons AI won't cite them.