How Often You Need to Update Content to Stay Cited by AI Search

A page can rank the same, read the same, and say nothing wrong, and still quietly stop showing up in ChatGPT and Perplexity answers a year after it was published. Not because it got outranked. Because it got old, in a way that classic SEO never punished this directly. A July 2026 study from Seer Interactive tracked 7,683 pages carrying 47,097 citations across ChatGPT, Gemini, and Perplexity, and found the freshness gap is real, measurable, and different by engine, which matters if you're deciding where to spend a content team's limited hours.

How an AI answer engine actually decides a page is "current"

An AI answer engine doesn't crawl once and treat every page as permanently ranked, the way classic Google search built its link-graph authority. Most answer engines break a query into smaller sub-questions (a "fan-out"), retrieve candidate passages for each one, and then have to decide, passage by passage, which ones are current enough to trust. That decision leans on machine-readable freshness signals more than a human reader would guess: schema.org's `dateModified` property, the `lastmod` tag in your XML sitemap, and the HTTP `Last-Modified` response header all exist specifically to answer "when was this actually last true," separately from when it was first published.

Google's own documentation is blunt about the limits of that signal: "Google uses the `lastmod` value if it's consistently and verifiably (for example by comparing to the last modification of the page) accurate." It also specifies what counts as significant enough to justify updating that date: "an update to the main content, the structured data, or links on the page," not a copyright-year bump. That distinction, a real content change versus a cosmetic date change, is exactly where the Seer data gets interesting.

What the data actually shows, and it's not the same number for every engine

Seer's study compared last-modified dates against citation frequency across three engines, using structured signals (schema, sitemaps, headers) pulled from real client sites in retail, banking, travel, and pet e-commerce between March and June 2026, against a June 2025 baseline. Overall, 75% of cited pages had been updated within the last year, 88% within two years, and citations from content older than three years were negligible. But the by-engine split is what actually changes how you'd plan a content calendar:

The more useful finding sits underneath that table: of the pages Seer could check both dates for, 72% looked fresh by last-modified date, but only 42% looked fresh by their original publish date. In their words, "the freshness LLMs reward is being manufactured by updates, not by new publishing." A page from 2022 that got a real edit last month reads as current. A page from last month that never gets touched again starts aging out within the year. That's the opposite of how most content teams are staffed, publish-heavy, update-light.

Freshness alone doesn't get you cited, structure still gates it

None of this works if the page can't be read cleanly in the first place. Peec AI's analysis of 30 million+ sources and 232,000 citations found citation share is extremely concentrated, on Perplexity, 64% of URLs never got cited at all, and pages with more than 60 navigation links delivered only 33% actual content on an AI agent's first read, versus 78% for pages with light navigation. Updating a bloated, nav-heavy page on a tight schedule won't fix that; the structure has to earn a citation before freshness can help it keep one. We covered the structural side of this separately in why schema markup alone doesn't move AI citations and what actually predicts brand citation, mentions versus backlinks, both worth reading alongside this if you're building a full GEO checklist rather than chasing one lever.

Don't fake the date, Google's own systems are built to catch it

The honest caveat here matters more than the growth-hacky version of this advice. Bumping a `dateModified` field or a sitemap's `lastmod` tag without an actual content change is exactly the abuse pattern Google's documentation is written to filter out, "if it's consistently and verifiably accurate" is a standard you fail the first time your claimed update date doesn't match what a crawler can independently confirm changed on the page. Once a domain's dates prove unreliable, the safest assumption is that the signal gets discounted for that domain going forward, not just for the one page. The Seer data backs the same conclusion from the other direction: the pages that held citations across all four months they measured averaged six months since their last real update, not six months since a metadata tweak.

A cadence you can actually run

Treat freshness as scheduled maintenance on your highest-value pages, not a blanket policy:

• Pages tied to a comparison, pricing, or tool landscape (the kind Gemini favors) need a real edit roughly every 4-8 weeks, a new data point, an updated price, a tool that entered or left the comparison, since these are the categories Seer found engines re-checking most aggressively.
• Blog-style guides ChatGPT tends to cite can run on a 3-4 month cycle: add a section, correct anything the underlying tool or platform changed, refresh a stat that's now a version behind.
• Reference and glossary-style pages Perplexity tolerates staying older can run every 6-12 months, evergreen content earns some slack, just not indefinite slack.
• Update the visible on-page date, the schema `dateModified`, and the sitemap `lastmod` together, in the same edit, not on separate schedules, since a mismatch between what the page shows and what the metadata claims is the inconsistency Google's own guidance flags.

Where this fits into a real GEO program

Freshness is one input, not the whole system, it sits alongside the mention-building and structural work we've covered in the posts linked above, and alongside actually measuring whether any of it is converting once you can see the traffic. Building and running that full stack, what to update, how often, how to structure a page an AI agent can actually parse, and how to prove it's working, is the operational core of The AI Marketing Engine: not a single tactic from a single study, but the content-ops system that keeps a site's highest-value pages actually current instead of quietly aging out of the answers your buyers are reading.