Most content teams treat AI citations as a publishing problem: write more, cover more ground, ship faster than the competition. A new Seer Interactive study on content recency says that instinct is backwards. Wil Reynolds' team found that 75% of pages cited by LLMs were updated within the last year, and 88% within two years — but when they checked original publish dates instead of update dates on those same pages, only 42% were actually written recently. The practical lesson for anyone trying to refresh content for AI citations: the winning move usually isn't a new post. It's an old one, done right, on a schedule.
Everyone is publishing more when they should be refreshing what already ranks
Ask most content teams how they plan to win AI citations and you'll get some version of the same answer: publish more, publish faster, cover the gaps competitors haven't gotten to yet. That answer made sense in a world where distribution rewarded net-new URLs. It's the wrong answer for AI search, where the signal engines are actually keying on is recency of the page itself, not its birthdate. Seer Interactive's Wil Reynolds published a study on July 24, 2026 that quantifies the gap between those two things for the first time at any real scale, and the gap is large enough to change how a content team should be spending its time.
The headline numbers look like validation for the publish-more instinct at first glance: 75% of pages cited across ChatGPT, Gemini, and Perplexity had been updated within the past year, and 88% within the past two years. Read quickly, that looks like AI engines have a hard bias toward brand-new content, which would justify throwing more budget at net-new production. Read carefully, using the second half of Seer's methodology, it says almost the opposite. The study also tracked original publish date separately from last-modified date, and that second number tells the real story.
The 72% freshness illusion: update date versus publish date
Here's the nuance that should reset how most teams read "freshness" as a ranking factor. Measured by update date, 72% of cited pages in Seer's sample looked fresh. Measured by original publish date, only 42% actually were. That's a 30-point gap, and it isn't noise — it means more than a quarter of the pages AI engines are treating as current were first published two-plus years ago and later refreshed, not written from scratch. The engines can't tell the difference between a page that's genuinely new and a page that's old but well maintained, and based on this data, they don't appear to care. What they're rewarding is evidence of recent attention, not evidence of recent authorship.
That 30-point gap is the entire argument for treating refresh work as its own discipline instead of an occasional cleanup pass. A team that only tracks publish velocity is optimizing for the 42% figure, chasing new URLs, while the engines are visibly rewarding the 72% figure, rewarding maintained ones. Every quarter spent purely on net-new production while a library of aging, still-relevant pages sits untouched is a quarter spent optimizing the metric that matters less. This is also where a lot of GEO strategy gets built on the wrong foundation: a team reads "AI engines favor fresh content" and assumes the fix is a faster editorial calendar, when the data says the fix is closer to routine maintenance discipline applied to what already exists. We've made a version of this case before in our breakdown of what actually drives AI citations across a billion data points: volume alone was never the lever. Recency signal is a different lever, and it points at old pages, not new ones.
How to refresh content for AI citations, by content type
The 42%-versus-72% gap isn't evenly distributed across a site. Seer broke out the share of cited pages that required freshness to earn a citation by content type, and the range runs from 78% down to 45%. That range is the difference between a page type that needs monthly attention and one that can go a year or more between touches without losing citation share.
| CONTENT TYPE | SHARE REQUIRING FRESHNESS TO GET CITED | WHAT THAT MEANS FOR CADENCE |
|---|---|---|
| Marketplaces | 78% | Near-constant refresh — pricing, inventory, and offers go stale within weeks |
| Comparison / reviews | 77% | Frequent refresh — competitors change roadmaps and pricing faster than most review cycles |
| Reference | 74% | Regular refresh — facts and figures drift even when the underlying topic doesn't |
| Brand / corporate | 72% | Regular refresh — leadership, positioning, and offerings change more often than corporate pages get touched |
| Blogs / guides | 67% | Moderate refresh — update on a fixed cycle, not only when something breaks |
| News / editorial | 45% | Longest runway — often cited for historical context, so age is less of a liability |
The pattern makes sense once you think about why an AI engine would care about recency at all: it's a proxy for accuracy. A marketplace page with stale pricing is actively wrong, so engines lean hardest on recency there — 78% of cited marketplace pages needed to look fresh to earn the citation. Comparison and review content sits close behind at 77%, for the same reason: a comparison built on last year's feature set is misleading a buyer, not just outdated. News and editorial content sits at the other end at 45%, and that's structural rather than a quality signal — a lot of news content gets cited specifically for what happened on a given date, so its age is the point, not a defect. Blogs and guides land in the middle at 67%, which is exactly the profile of content most teams treat as "evergreen" and then genuinely forget about. If your comparison pages haven't been touched since launch, our research on why comparison content wins a disproportionate share of citations is worth reading alongside this table, because the format that wins citations most often is also the format with the least tolerance for going stale.
Which AI engines reward a content refresh for AI search the most
Content type is one axis. The AI engine your buyers actually use is the other, and Seer's engine-level breakdown shows real spread: Gemini cited content updated within the last year 78% of the time, ChatGPT did the same at 73%, and Perplexity trailed at 65%. Perplexity is meaningfully more tolerant of older content than either of the other two, which matters a lot if you're building a refresh program against a fixed budget and need to decide where the marginal hour goes.
Share of cited content updated within the last year, by AI engine (Seer Interactive, Jul 2026)
If Gemini or ChatGPT visibility matters most to your pipeline, this is not optional maintenance — it's close to a prerequisite. A page that hasn't been touched in eighteen months is fighting recency math that roughly three in four competing citations are already winning. If Perplexity carries more of your buyer intent, the same page has a longer runway before staleness becomes the deciding factor, though "longer" is not "forever" — 65% is still a majority, not a rounding error. Either way, a refresh program that ignores which engine matters to a given audience is guessing at a variable Seer's data actually answers. Pair the engine you're optimizing for with the content type table above and you get a real prioritization matrix instead of a flat "update everything quarterly" rule that wastes effort on low-freshness-need pages while under-serving the ones that need monthly attention.
A refresh cadence you can actually run
None of this works as a one-time audit. The Seer numbers describe an ongoing state of the AI-cited web, which means a refresh cadence has to be a standing operating rhythm, not a project you run once and mark complete. Here's the sequence we'd run against the data above, from fastest cadence to slowest.
This same logic scales differently depending on the size of the catalog you're maintaining. A B2B site with a few hundred pages can run this cadence manually with a shared spreadsheet and a calendar reminder. An ecommerce catalog with tens of thousands of product and category pages needs the same principle applied at a different scale — freshness signals concentrated on the highest-intent marketplace and comparison pages first, since that's where Seer's data shows the steepest freshness requirement and the highest citation payoff; see how that plays out in practice in our work with ecommerce catalogs at scale. Either way, the sequencing matters more than the tooling: inventory first, cadence by type second, cadence by engine third, substantive-update discipline fourth. Skip step one and you'll refresh pages that didn't need it while the truly overdue ones stay untouched.
“The 30-point gap between 72% and 42% is the whole strategy in one number: AI engines are telling you, in aggregate, that they'd rather cite a well-maintained old page than an untouched one that happens to be new.”
Do this next: pull your ten highest-traffic or highest-citation-potential pages this week, tag each by content type, and check the last substantive update date against the table above. Any marketplace or comparison page untouched in the last quarter is your first refresh. Anything else can wait for the cadence you set in step two. That's a smaller, more specific task than "publish more content," and based on what Seer's data shows about how AI engines actually treat recency, it's the one more likely to move a citation number in the next reporting cycle.
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.