The three freshness channels
- Declared dates. datePublished and dateModified in Article schema, mirrored visibly on the page. Grounding systems use these to weight and to attribute ("as of…"). Missing dates don't disqualify you — they just make you undateable, which for fast-moving topics is nearly the same thing.
- Retrieval recency. Live-retrieval engines (Perplexity above all) weight recent sources for evolving questions. A pricing page updated last month beats a competitor's from 2024 for "current pricing" queries — and the engine will say "as of" the date it finds.
- Crawl priority. Sites that publish real updates on a cadence get re-crawled more often. Freshness compounds: maintained sites stay closer to the front of every retrieval pool.
Update cadence that works
- Money pages: review monthly; update when facts change, stamp honestly.
- Benchmark/data content: quarterly editions beat continuous silent edits — citable artifacts with versions (our State of GEO pattern).
- Reference content: annual review is fine if dates are honest; evergreen-with-a-date beats fake-fresh.
The honesty rule
Do not bump dates without changing content. The pattern is detectable (dateModified churn with byte-identical content) and teaches engines to discount your declared dates entirely — you spend your most useful trust signal for nothing. Update the date when you update the substance; both or neither.
Frequently asked questions
How often should I update content for AI visibility?
On the cadence your facts actually change: monthly review for money pages, quarterly for data pieces, annual for evergreen reference. Real updates, honestly stamped. Cadence theater (date bumps without substance) actively damages the signal.
Do engines penalize old content?
Not for age alone — for staleness. A 2023 page with accurate 2026-relevant facts and an honest review date performs fine; a 2026-dated page with 2023 facts performs badly once contradicted by fresher sources. Accuracy is the substance; dates are the wrapper.