“One really good article and you’re set — it just sits there working for you.” Every conversation about content gets to this line eventually, doesn’t it?
I feel it too. An old post I sweated over still drags people in from search, and it feels like free money — one afternoon of work, paying out for years.
But lately I’ve been watching which pages the AI actually cites when someone asks it for a recommendation, and something’s off. The page that showed up all the time last year has quietly been swapped for some newer thing I’d never seen.
Somebody actually measured this “the AI won’t cite my old pages” problem across more than 5,000 pages. So let’s look at what they found.
Why “Publish Once” Stopped Working In AI Search
In (search engine optimization, the craft of ranking high on Google), a page that clawed its way to the top tended to stay there a while. That’s the whole reason “publish it and move on” ever made sense.
But the pages AI pulls into its answers are twitchier. Ask the same question a month apart and the lineup of cited pages shifts. Less a fixed bookshelf, more a new-releases table someone restocks every month.
So the real question is: how new does a page have to be before the AI likes it? Somebody put a number on it.
What 5,000 Pages Say About Which Ones Get Cited
The study comes from Seer Interactive, a US digital marketing firm (R). They gathered more than 5,000 pages with known publish dates and counted, across ChatGPT, Perplexity, and AI Overviews (the AI summary Google now bolts on top of search), which years’ articles were getting cited.
The method’s refreshingly grubby. They followed the access logs — the record of AI bots actually crawling a site — to see which pages got “picked up,” then cross-checked that against citation data. So this is real hit counts, not vibes.
One caveat worth keeping in your back pocket: Seer also sells a content-refresh service. So “and that’s why you should update your content” is a conclusion they’re paid to reach. But the raw trend in the data is hard to argue with.
65% From The Last Year, And Just 6% Beyond Five Years
Here’s what came out. The AI bot hits leaned hard — almost comically hard — toward recent pages.
- About 65% of all hits went to pages published in the last year.
- 94% went to pages from the last five years — which means everything older than five years splits a measly 6%.
So that killer piece you wrote six years ago is, as far as the AI’s concerned, basically not there. A page that might still rank fine in SEO gets dropped cold from AI citations. I think that’s the finding that actually matters.
Perplexity Is The Fussiest, And Your Industry Matters Too
It gets better. The models have different tastes. Of the pages they cited, the share published this year was 31% for ChatGPT — but a full 50% for Perplexity. Perplexity, it turns out, is the one that really only wants the brand-new stuff.
Your industry changes the math too. Financial-services content lives and dies by recency, while how-to pages — the “here are the steps” kind — keep getting cited long after they’re old. So which side your topics fall on decides how hard you lean on updating.
Conclusion: Go Check The “Last Updated” Date On Your Big Pages
Put it together and the takeaway is this: visibility in AI search isn’t a build-once asset. It’s something you hold onto by staying fresh. (generative engine optimization, SEO for the answer ChatGPT gives instead of the links Google shows) doesn’t reward set-it-and-forget-it the way SEO sometimes did.
So today’s move is small. Pick a handful of pages that matter to you and just look at when they were last updated. If a flagship page hasn’t been touched in years, there’s a decent chance it’s already sliding off the AI’s list of things worth citing.
Then start nudging your calendar. Less “published, on to the next post,” more “revisit the important pages I already have.” You don’t need to chase every URL — even giving the central pages for your key topics a real once-over every six months would move the needle.
I’ve started doing my own version of this: pulling up old workhorse posts and asking, “is this still true?” For AI-era content, I think that one small habit is the most realistic place to start.
