There's no fixed date — what matters is that there are real changes and that they're declared: dateModified in the schema of every page that actually gets updated, and a sitemap with a real lastmod. The gap shows up clearly: a University of Toronto study measured that the three leading AIs cite content between 30% and 52% fresher than what a traditional search engine offers for the same question.
On consumer electronics, the same study measured the real age of cited content: Claude cites articles with a median age of 62 days, GPT-4o 80, and Perplexity 90 — versus 130 days for Google searching the same question.
For automotive, the same study measured an even bigger gap: median cited age was 148 days (Claude), 162 (GPT-4o) and 217 (Perplexity) — versus 493 days for Google. The proportion holds, even though the whole industry moves slower: a car's information changes less often than a phone's, and yet the AI advantage over Google widens instead of shrinking.
The study doesn't give a single cause, but it leaves a clear clue: in both industries, Claude and GPT-4o concentrate their citations on "earned" media — specialized press and independent review sites, not official brand pages — like TechRadar, Tom's Guide, RTINGS and CNET for electronics, or Consumer Reports and Car and Driver for automotive. That type of source republishes and updates content far more often than a static product page. The pool an AI can cite as "the latest" is already skewed toward what gets updated often.
Unlike other criteria, here Nostos doesn't score how "new" the text sounds — that isn't verifiable — but structured update signals: the dateModified field declared in the page's schema, and the presence and validity of a sitemap.xml with real lastmod dates. A site can read as up to date and declare none of this; to an AI, without the structured declaration, that update doesn't exist.
Nostos's own site measured this firsthand. Before having a sitemap.xml, its Freshness and updates score was 68/100. On 07/02/2026 a sitemap was added with the real publication dates of each URL — without writing a single line of new content, just declaring in a machine-readable way when each page had last changed. The score rose to 82/100 in the next measurement.
Those 14 points didn't come from new content — they came from declaring, in a way a machine can read, information the site already had.
sitemap.xml with real lastmod dates, that's first — it's cheap to implement and Nostos's own case showed a direct 14-point jump.dateModified in the schema of every page that actually gets updated, not just its original creation date.An AI doesn't cite you because you claim to be up to date — it cites you if it can verify it. The declared date and the real content have to say the same thing.
Get your GEO score across the eight criteria backed by academic research, including your current Freshness and updates.
Run the analysis at nostosgeo.app