Seer Interactive published a study on July 24 examining how content age relates to what AI models cite. The agency analyzed 7,683 dateable pages carrying 47,097 citations in non-branded ChatGPT, Gemini, and Perplexity answers between March and June 2026, drawn from four industries: a national pet retailer, a vacation rental marketplace, a retail energy provider, and a commercial bank. Three quarters of cited pages had been updated within the past year, and 88% within the past two, per the study by Seer's Sonny Vasquez.

The sharper finding is where the freshness comes from. Of the cited pages, 72% look fresh by their last update, but only 42% were published recently, and more than a quarter of the fresh-looking pages were originally published two or more years ago. Older pages that get refreshed earn more citations than genuinely new ones. "Publish and forget loses. Publish and maintain wins," Vasquez writes.

The three models weigh recency differently. Gemini cites the freshest material, with 78% of its cited pages updated within a year. ChatGPT sits at 73%, and blogs and guides account for 46% of what it cites. Perplexity is the most tolerant of older reference content at 65%. The study also separates citation spikes from durable presence: pages cited in a single one-month burst skew heavily fresh, at 86%, while always-on pages cited in every month of the four-month window are 68% fresh with a median age around six months. Staying cited every month, the study argues, is a different game from catching a spike.

The caveats are real. This is agency research on Seer's own client set, so the industry mix is narrow, and only about two thirds of pages could be reliably dated; treat the percentages as vendor-reported and directional. A companion Seer study looks at the same recency question from the brand-visibility side.

The takeaway for brands: AI models are disproportionately reading and citing pages that were recently touched, so the update cadence on your existing high-value pages is doing more for how models see you than net-new publishing. The cheapest move this study points at is a maintenance calendar: keep the pages models actually read current, and treat their upkeep as seriously as the next launch post.