Does updating old content get you cited by AI?

The standard advice is to keep publishing: a new post every week, a steady stream of fresh URLs. A large citation study published this July says the businesses winning AI citations are mostly doing something cheaper — going back and fixing pages they already have.

Seer Interactive pulled every page cited in non-branded answers for four brands across four categories — a national pet retailer, a vacation rental marketplace, a retail energy provider, and a commercial bank — across ChatGPT, Gemini and Perplexity over a shared four-month window, March through June 2026. For each cited page they pulled a last-modified date from schema, sitemaps or headers, and could date about two thirds of them: 7,683 pages behind 47,097 citations (Seer Interactive).

The twist: refreshed beats new

So far that just confirms what everyone assumes — fresh wins. The useful part is what Seer added this year: they separately checked each page's original publish date against its last-update date, on the 4,124 pages where both were readable.

In the researchers' words: "More than a quarter of the 'fresh' pages were first published over two years ago. They earned their freshness by being maintained." The pattern held in every vertical studied, with the update-vs-publish gap running 22 to 35 percentage points depending on the category. A page you wrote two or three years ago and genuinely revised last month reads as current to the model. A page you wrote last month and never touch again starts aging the day you hit publish.

What this means if you are not publishing every week

Most small businesses do not have the staff for a constant content calendar, and this data suggests that may not be the bottleneck it looks like. Blogs and guides made up half of all cited pages in the study and are the content type Seer calls an "owned lever" — yours to update directly, unlike marketplace listings or comparison sites you do not control. Brand and corporate pages, the smaller but fully-owned category, were updated within the year at 72%, right at the study average.

Which engine rewards this most matters for where to spend the afternoon. ChatGPT leans hardest on exactly this content type: "blogs and guides are 46% of its citations, nearly half," and it cites brand and corporate pages more than Gemini or Perplexity do (10%). Perplexity is the most forgiving of the three, still pulling from older evergreen material others skip. Gemini is the freshest overall, but skews toward marketplace and comparison content most small businesses don't have — "if your own pages are going to earn citations anywhere, ChatGPT is the most likely place, so this is where owned-content freshness matters most."

The practical version: instead of writing a new page nobody asked for, open the handful of pages that already answer a real customer question — pricing, service area, how the process works, the most-asked question in your inbox — and change something real in each: this year's number, a corrected detail, a current example, a rewritten intro paragraph. Then update the visible date. Consistency matters more than speed here too: pages cited across all four months of the study were 68% fresh with a median update age around six months, not bleeding-edge — a page revisited every couple of quarters clears that bar.

A second study, a different method, same direction

An independent paper audited 1,100 URLs behind 1,702 citations across Brave Summary, Google AI Overviews and Perplexity, scoring each page against 16 quality "pillars." Of all sixteen, Metadata & Freshness correlated most strongly with getting cited (r = 0.68) — ahead of Semantic HTML (r = 0.65) and Structured Data (r = 0.63) (arXiv — GEO-16 framework). That last one is worth sitting with: structured data outranked by freshness is the same conclusion Ahrefs reached testing schema markup directly — the technical fix everyone reaches for first moves less than simply keeping the page current.

Two different methods, two different datasets, the same signal near the top both times. Neither proves that updating a page causes it to get cited — Seer's data is descriptive, and a correlation of 0.68 is strong but not proof. What both studies support is where the odds sit: closer to maintenance than to volume.