Freshness: the citation signal nobody budgets for

Content budgets fund new pages. The citation data says a large share of that money belongs in old ones. AI engines show a strong, measurable preference for recently updated content, and almost every content plan ignores it, because maintenance is nobody's deliverable.

What the citation data shows

A study across 7,683 pages and 47,097 citations measured how fresh the content cited by AI engines actually was. Gemini cited content that had been updated within the year 78 percent of the time. For ChatGPT the figure was 73 percent, and for Perplexity 65 percent. Yet only 42 percent of cited content was recently published.

The gap between those numbers is the finding. If engines simply favored new pages, the updated share and the recently published share would match. They do not, and the difference describes the ideal citation target: an established URL that has been genuinely refreshed. New pages start with no history; stale pages fall out of the 65 to 78 percent that engines prefer. The page that wins is the one that has existed for years and was updated this year.

What counts as updated, credibly

The tempting shortcut is the date-stamp game: change the visible date, touch a sentence, republish. That produces a page that claims to be current while containing nothing current, and it misses how citation works. An engine grounding an answer lifts concrete material, so a refresh only matters if it changes the substance a model would actually quote:

• Current figures. Replace last year's statistics with this year's, from named sources.
• Current sources. Swap superseded reports and dead links for the latest versions.
• Revised conclusions. Where the market moved, say what changed, not just when.
• Removed claims. Delete what no longer holds instead of leaving it to be checked against reality.

This is the same substance the GEO research found effective in the first place: adding statistics lifted visibility in generated answers by around 30 percent and citing sources by around 27 percent. A real refresh reapplies those interventions with current material. A cosmetic one adds nothing a model can use.

A refresh workflow

1. Find the decaying pages. In Search Console, compare the last twelve months of clicks and impressions against the twelve before. Pages that once earned traffic and are sliding are your refresh queue; they have the history new pages lack and the staleness engines discount.
2. Rewrite against current sources. Re-research the topic before touching the copy. If every number and reference survives unchanged, the page did not need a refresh and should not pretend it got one.
3. Signal the update honestly. Update the visible date only when the substance changed, and keep the lastmod value in your sitemap accurate. Google's sitemap documentation says it uses lastmod when it is consistently and verifiably accurate, which means inflating it teaches Google to ignore yours.

Freshness is a loop, not a project

A refresh done once decays like everything else, so the workflow above belongs on a calendar, with the queue rebuilt from Search Console each cycle. Whether it works shows up as recovered impressions and, on a slower clock, as citations returning in your measurement rounds. Vupie runs this refresh loop on autopilot for its customers, but nothing in it requires tooling, only the discipline to fund maintenance. What the refreshed substance should look like to earn the citation in the first place is covered in how to get cited by AI engines.