Here is a finding that surprised us more than it should have: across our client audit set, pages that show a visible "last updated" date were cited roughly 2.3x more often by Perplexity than pages without one — holding the actual content constant. Same words, same author, same topic. The only difference was a small line of text near the top that said when the page was last touched.
That is a large effect for an almost-free change. This piece explains why generative engines lean on freshness so heavily, how to surface the signal so a model actually sees it, and — importantly — how to do it honestly, because the shortcut version backfires.
Why do LLMs reward freshness?
Generative engines reward visible freshness because recency is the cheapest available proxy for reliability, and stale answers are one of their most damaging failure modes. A model that confidently cites a 2019 statistic as current looks broken to users. So when an engine retrieves candidate passages, a clear, recent date is a thumb on the scale toward trusting — and quoting — that source.
There is a structural reason too. Every model has a training cutoff, and retrieval-augmented systems bolt live search on top precisely to cover what happened after that cutoff. In that architecture, recent content is doing the job the model cannot do from memory. Freshness is not a vanity metric to these systems; it is the entire reason the retrieval layer exists.
A model would rather cite a fresh, decent answer than a stale, excellent one — because it cannot tell from the text alone whether the excellent answer is still true.
How much does a visible date actually move citation?
In our six-month audit window, the visible-date effect clustered around 2.3x for Perplexity and was meaningful but smaller for Google AI Overviews. The mechanism appears to be selection, not ranking: among several passages that could answer a query, the engine disproportionately lifted the one whose page advertised recency.
The date has to be where the model reads
Placement matters more than people expect. A date that exists only in your XML sitemap or in a meta tag is weakly weighted; engines parse the rendered page, and a human-visible "Last updated: …" line in or near the article body is far stronger. If your CMS hides the date in a tooltip or a machine-only field, you are leaving most of the effect on the table.
Pair the date with a real change
The effect compounds when the visible date coincides with genuinely updated content — a refreshed statistic, a new paragraph, a corrected claim. Engines increasingly diff content over time; a date that moves while the words sit still is a pattern they can learn to discount.
Does faking the date work?
Briefly, then it backfires. Restamping a stale page with today's date can produce a short-term citation bump, but generative engines cross-check claims against the actual content and against other sources. A page that claims freshness while repeating outdated facts is exactly the failure case these systems are built to avoid — and once a source is associated with that pattern, it tends to lose citation broadly, not just on the gamed page.
We have watched this happen on competitor pages in client audits: a flurry of "updated today" stamps across an entire blog, a brief lift, then a slide as the engines stopped trusting the signal from that domain. The juice is not worth it. Treat the date as a promise you keep.
How to build an honest freshness cadence
The durable version of this is a real refresh routine on the pages that matter, with the visible date reflecting genuine work. Here is the cadence we run for clients, and that you can run by hand.
- Identify your top 20 pages by traffic and by buyer intent — not the whole site, just the ones that earn citations.
- Surface an honest, human-visible "Last updated" line near the top of each.
- On a monthly cycle, revisit each: refresh one statistic, tighten the lead paragraph, add one new fact. Then move the date.
- Log what changed, so the date is always backed by a real diff you could defend.
- Re-run an LLM citation audit after each cycle to confirm the engines noticed.
This is exactly the loop Avakata automates: the engine watches the citation feed, drafts the smallest honest refresh that restores a slipping page, runs it past a critic, and ships it — then moves the visible date because the content genuinely changed. The freshness is real; only the labor is removed. See how citation and freshness fit together in [Why citation is the new ranking](/blog/citation-is-the-new-ranking), or [book a discovery](/contact.html) to watch it run on your pages.