A study came out in late July that is about to be used to justify a lot of bad content budgets. I want to get to it before your Q4 planning does, because the headline number and the number three rows below it point in completely opposite directions, and almost everyone is quoting the first one.
Sonny Vasquez at Seer Interactive published the analysis on July 24. The sample is real: 7,683 pages carrying 47,097 citations across ChatGPT, Gemini and Perplexity, drawn from non-branded answers for four brands in pet retail, vacation rentals, energy and commercial banking, between March and June 2026. Last-modified dates were pulled from schema, sitemaps and headers, and Seer is upfront that only about two-thirds of pages could be dated successfully.
The headline that traveled: 75% of pages that get cited were updated within the last year, 88% within two. More than half of the recently-updated group had been touched within the previous three months. By engine, Gemini leaned hardest on recency at 78% within a year and 90% within two, ChatGPT sat at 73% and 87%, Perplexity was the most forgiving at 65% and 83%.
Every one of those numbers is being read as a mandate to publish more. It is not. It is a mandate to maintain more, and the difference between those two sentences is most of a content budget.
The study everyone quoted half of
Here is the row that changes the meaning. Seer measured freshness two ways. By last-update date, 72% of cited pages look fresh. By publish date, that drops to 42%.
Share of AI-cited pages that read as recent, measured two ways (Seer, 7,683 pages, Mar-Jun 2026)
Thirty points of the cited population are pages that were published a while ago and updated since. Not new content. Old content, maintained. That is the largest single actionable group in the dataset, and it is the one that costs the least to produce, because you already produced it.
What content recency and AI citations actually means
Corroborating data points the same way. Kevin Indig and Amanda Johnson's analysis of roughly 35,000 citation URLs found 83% of AI citations for commercial queries came from pages updated in the past twelve months, and that stale pages were more than three times as likely to lose AI citations. Meltwater's research, cited via AMEC in July, put 58% of cited content at under a year old. Three independent datasets, one direction.
But notice what none of them say. None of them say a page has to be new. They say the engine's picture of a page has to be current. Those are different requirements, and only one of them requires you to keep hiring writers to produce things you do not yet have.
There is a per-engine wrinkle worth planning around too. Perplexity at 65% within a year is meaningfully more tolerant of older material than Gemini at 78%. If your buyers concentrate on one engine, your maintenance cadence can follow that engine's tolerance rather than the industry average. Most teams will never segment this finely. The ones who do will spend less to hold the same ground.
The refresh case, in money
Run the arithmetic on a mid-sized content operation and the argument stops being philosophical. Take a library of 400 published articles and a budget that currently produces twelve new pieces a quarter. Reallocating a third of that capacity to refresh work — same writers, same hours — covers roughly forty existing pages per quarter at the effort ratio we typically see, because updating a page you own is a fraction of the work of researching one you do not.
| APPROACH | QUARTERLY OUTPUT | PAGES MADE CURRENT | WHAT YOU ARE BETTING ON |
|---|---|---|---|
| All net-new | 12 new articles | 12 | That twelve new pages out-earn the decay of 400 existing ones |
| Two-thirds new, one-third refresh | 8 new + ~40 refreshed | 48 | That currency across the library beats volume at the edge of it |
| Refresh-led with selective new | 4 new + ~80 refreshed | 84 | That your existing library already covers the demand and needs maintenance, not expansion |
Those output ratios are illustrative and will vary with how deep your refreshes go — a fact-check and data update is not the same job as a structural rewrite. The point is the shape. Under any reasonable effort assumption, a refresh hour touches more of your citation surface than a net-new hour, and the Seer data says the citation surface is disproportionately made of maintained pages.
“You do not have a content production problem. You have four hundred pages that were true when you wrote them.”
Three ways teams will get this wrong
Content recency is a proxy, not a lever
Now the part that keeps this honest. Every study here is correlational, and there is a very plausible confound sitting in plain sight: the pages that get maintained are the pages someone cares about. Caring shows up as accuracy, better sourcing, cleaner structure, working links and current examples. Recency might be the visible fingerprint of quality rather than an independent ranking input.
That is not a reason to ignore the finding. It is a reason to act on it in the right direction. If recency is a proxy for maintenance and maintenance is a proxy for quality, then the intervention that works is doing the maintenance, and the intervention that fails is manipulating the date. Which is exactly what the two readings of this study recommend, and exactly why the difference matters.
Two more limits to keep in view. Seer dated only about two-thirds of the pages in its sample, so the freshness distribution describes datable pages rather than all cited pages. And four brands in four categories is a real sample, not a universal one — a category where the underlying facts move slowly may behave differently from commercial banking. Treat the numbers as a strong prior, then verify against your own citation set.
Running a refresh program that works
The operating model is simple enough to run with existing people, and it is mostly a prioritization exercise rather than a creative one.
None of this argues for stopping new content. New pages are how you enter conversations you are not in, and if your library genuinely does not cover a question your buyers ask, no amount of refreshing will fix that. The argument is about ratio, and the ratio in most content plans we review is set by habit rather than by evidence. The habit is understandable — new pieces are easier to report on, easier to celebrate, and easier to point at in a quarterly review than forty quiet corrections spread across a library nobody outside the team has read end to end.
There is an organizational cost to fixing this and it is worth naming. Refresh work is invisible in most content dashboards, which count publications. If your team is measured on output, asking them to spend a third of their capacity on work that shows up nowhere is asking them to look less productive on purpose. Change the metric first, or the program will quietly revert within two quarters. Put refreshed-pages and persistence on the same slide as published-pages, give them equal visual weight, and the behavior follows. This is the same reporting discipline that makes AI visibility measurement credible rather than decorative.
Do this next
Pull the last-update dates for your top hundred pages this week and sort ascending. If more than half were last touched over a year ago, you have found the cheapest available improvement to your AI visibility, and it does not require a single new brief.
Then take the first ten and give them a real hour each: current numbers, live sources, sharpened headings, honest date. Do that every quarter and you will out-cite teams publishing three times your volume. Our content marketing work is built around this maintenance cadence rather than a publishing quota, and the supporting arguments sit in our guide to content that earns AI citations and our case for depth over breadth in topical authority. Seer published the full recency study with its per-engine breakdowns if you want the underlying tables.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.