Yes, and the date that matters is the last update, not the day you published. Seer Interactive dated 7,683 pages that ChatGPT, Gemini and Perplexity cited across 47,097 citations between March and June 2026 and found that 75% had been updated in the past year, while only 42% had been published in it. Ahrefs reached a similar conclusion a year earlier from a larger and blunter sample: across 16.975 million cited URLs, the average page an AI assistant cites is 1,064 days old, against 1,432 days for the pages in Google’s organic top ten, which makes AI citations roughly a year fresher than the search results they are replacing. The practical answer for a mid-market company is that the article you wrote in 2023 is still your best asset, provided somebody has touched it since.
What the two studies actually measured
The two datasets answer slightly different questions, and the difference is where the useful advice lives. Ahrefs used Brand Radar to pull every URL cited across ChatGPT, Perplexity, Gemini, Copilot and Google’s AI Overviews, then compared the age of those pages with the age of pages ranking organically for the same queries. The AI assistants prefer newer pages, but not uniformly. ChatGPT’s in-answer citations averaged 958 days old, Copilot 1,056, Gemini 1,118 and Perplexity 1,166. Google’s AI Overviews cited pages that averaged 1,432 days, sixteen days older than the organic results beneath them, which is a polite way of saying that the Overview is the search index wearing a different hat.
Seer’s July 2026 study is smaller, four brands in four categories, but it did something Ahrefs did not: it separated the date a page was first published from the date it was last modified, using schema, sitemaps and headers. Measured by last update, 72% of cited pages looked fresh. Measured by publish date, only 42% did. More than a quarter of the pages the models treated as current had first appeared more than two years earlier. The freshness the machines reward is manufactured by maintenance, and the pages the models rely on most, the ones cited every month of the study, were 75% updated in the past year but only 40% published in it.
Why a refreshed page beats a new one
There is a mechanical reason and a commercial one. Mechanically, a page that has been cited for a while has already been crawled, chunked and associated with a set of questions in whatever the model’s retrieval layer is calling its index this month. Updating it changes the content without resetting the associations. A new page has to earn all of that from nothing. Seer’s persistence data shows the shape of it: pages cited in only one month of the four were 86% fresh with a median update age of about two months, while pages cited in all four months were 68% fresh with a median age of six months. A very recent update wins the first pickup. An established page that is kept reasonably current wins the repeat citation, and repeat citation is the only kind that changes a pipeline.
Commercially, most mid-market companies already own the raw material. Somewhere on the site is an explainer from 2022 that still ranks, a comparison page with two competitors who have since been acquired, and a pricing FAQ that quotes a tier you retired. Each of those is a candidate for the kind of freshness the machines reward, at a fraction of the cost of new production. Ahrefs’ own caveat is worth repeating: the average cited page is still 2.9 years old, the models are not chasing novelty, and updating a bad page frequently produces a frequently updated bad page.
The obvious shortcut does not work. Google’s John Mueller has said for years that changing the date on a page without changing the content is, in his word, noise, and that a date should move only when something new is written or something existing is significantly changed. The AI platforms read the same structured signals Seer used to date pages: schema, sitemaps, headers. Faking the date is the content equivalent of turning the clocks forward and expecting to be paid for the extra hour.
The machines reward the page somebody still looks after, which is a different and rarer thing than the newest page.
Where freshness pays and where it cannot
Seer sorted cited pages by type and by who controls them, and the sorting is the strategy. Marketplaces and aggregators were the freshest category cited, 78% updated in the past year, followed by comparison and review pages at 77%. Both are earned: you can influence a listing, but you cannot refresh someone else’s page on a Tuesday afternoon. Brand and corporate pages were 72% fresh and blogs and guides 67%, and those are the levers a company owns outright. Blogs and guides were also half of everything cited, and the category ChatGPT leans on hardest, at 46% of its citations. If your own pages are going to be cited anywhere, ChatGPT is the most likely place, and your guides are the most likely pages.
The engines differ enough to matter. Gemini was the freshest, with 78% of its cited pages updated in the past year, because it leans on marketplaces and comparison sites. Perplexity was the most forgiving at 65%, reaching further back for evergreen reference material. So a company with a deep, well-maintained archive has a natural advantage in Perplexity and a maintenance job to do for Gemini, where the freshness is on pages it does not own. That is the same lesson the Signal has been circling for months from other directions: review sites and comparison pages are where the citations live, and your own site is the smallest part of your own visibility.
What you can actually do this quarter
Find the twenty pages that are still cited
Pull your AI citation data, or failing that your top organic pages older than eighteen months. Those are the assets with associations already built. Nobody on the team has looked at them since they were published.
Set a refresh trigger, not a calendar
Seer’s advice is to let citation data set the threshold: a page that keeps appearing gets checked when its citations slip, not on an arbitrary quarterly rota. Three questions per page: have the facts held, is it still cited at the old rate, would being wrong cost anything.
Let the content type set the pace
Comparison and pricing pages need to be current, because the cited versions are. Evergreen explainers tolerate age. Do not spend the refresh budget on the pages that least need it, which is what a calendar would do.
Audit the pages you do not own
Your entries on review sites, directories and comparison articles have dates too. A listing last touched in 2023 is a freshness problem you can fix with a form and twenty minutes. Nobody has been assigned to it.
The part nobody wants to hear
Maintenance has no launch. Publishing a new piece produces a link to share, a post for the founder, a line in the monthly report. Rewriting the third paragraph of a 2022 guide because a statistic moved produces nothing anyone will notice on the day, which is why in most companies it is nobody’s job. The Spring 2026 CMO Survey found training budgets down to 3.8% of marketing spend and the most common capability shortfall was not a missing skill but a lack of people, time and budget to make existing capabilities work. That is the refresh programme, described by 308 marketing leaders who do not have one.
So the freshness advantage goes to the company that treats its archive as a portfolio rather than a feed: a fixed list of pages that matter, a person who owns each one, a trigger for when it gets touched, and a report that counts citations retained rather than posts shipped. Machines for scale, never for soul, applies here in a very literal way: the tooling can tell you which page slipped, but deciding what is still true about your business is a judgment, and judgment is the part that keeps getting deferred until the quarter is over.
The page that gets cited is the page somebody kept true, and keeping things true is a habit no tool can install.
The wider picture on getting cited by AI engines is in our answer engine optimization primer. If you would rather hand the whole problem to someone, see AI consulting for Connecticut companies.