Why stale pages vanish from AI answers

Here’s a pattern from our citation logs that changed how we run content programs. A client’s pricing guide sat in Perplexity’s citations for one of their top buy-intent queries for months. Then it aged past roughly a year old, and over a few weeks of checks it faded, replaced by a thinner page from a competitor, published recently. Same query. The better page lost to the newer one.

We’ve now watched versions of this happen across enough domains to call it what it is: AI engines have a freshness bias strong enough to override content quality at the margin, especially on anything time-flavored, prices, comparisons, “best,” anything with a year in it or implied.

Why engines over-weight recency

Put yourself in the model’s position. It’s assembling an answer about costs in 2026. It cannot verify which source is correct, but it can see which is current. Recency is the cheapest available proxy for reliability, so retrieval pipelines lean on it, Perplexity most visibly, AI Overviews clearly on time-sensitive queries, ChatGPT’s browsing mode in between. The engines aren’t wrong to do this. Stale pricing pages are genuinely worse answers. The bias is crude, but it points at something real.

What “fresh” means mechanically

Three signals, in rough order of weight as far as we can observe: visible content changes (engines compare crawl snapshots, cosmetic edits don’t register), honest dateModified in your Article schema, and visible “Updated” dates on the page. The order matters: the date stamps amplify real changes, they don’t substitute for them. Bumping dates on unchanged content is detectable and, from what we’ve seen, quietly punished.

The maintenance cadence we run

• Quarterly, for citation-critical pages: the 10–20 pages backing your priority queries get a real revision every quarter, updated numbers, a new example, current-year framing, refreshed dates. Plan an hour or two per page; it’s maintenance, not authorship.
• Annually, for the supporting tier: everything else in the cluster gets an honest once-over, accuracy check, dead links, one improved section.
• On trigger, for anything market-sensitive: price changes, competitor launches, regulation. The page updates the week the world changes, because that’s exactly when query volume and citation reshuffling spike together.

Freshness as moat, not chore

The optimistic reframe: this bias punishes set-and-forget content farms and rewards anyone willing to maintain a small set of genuinely current pages. A 20-page cluster, each page touched quarterly with real updates, will outperform a 200-page archive at citation acquisition, and it’s vastly easier to keep honest. Freshness is the recurring work in the GEO loop; pair it with weekly citation checks and you can literally watch a refresh land: the page re-enters rotation within days on fast engines. Few things in this discipline are that satisfying to observe.