Intro
Two pages can carry the same facts and the same authority. The one published last month gets cited; the one published three years ago does not. That is not a quirk — it is a measurable behaviour of several AI answer engines, and it changes how premium brands need to think about content maintenance.
This page explains what counts as fresh inside an AI engine’s selection logic, why fake dates fail under retest, and how a sustainable refresh cadence becomes part of the Capston Core methodology.
Why freshness matters more in AI than classical SEO
Classical SEO treats freshness as a soft ranking factor — useful for news and trending queries, secondary everywhere else. Generative answer engines do not behave the same way.
Chen, Wang, Chen and Koudas (2025) tested how five answer engines select sources across thousands of queries. Freshness was one of four dimensions where engine behaviour diverged sharply. Perplexity, in particular, leans heavily into recently-published content, often preferring a newer page over an older one with stronger authority. ChatGPT and Gemini are more tolerant of older sources when the domain carries weight. Google AI Overviews sit in between.
The practical consequence: the same brand page can be cited on one engine and ignored on another, purely because the engine reads its dates differently. A refresh cadence is no longer a content-team nice-to-have. It is part of the citation surface.
For premium brands, the asymmetry is sharper than for news publishers. A hotel does not republish a positioning page every week. But if that page carries a 2022 publish date and no modified date, Perplexity may treat it as stale and reach for a third-party source — an OTA, an aggregator, a review platform — that updates more often. The brand loses the answer not on substance, but on signal age.
What counts as fresh
Freshness is read by engines through several signals, not one.
- Publish date — the original
datePublishedvalue, visible in JSON-LD and often surfaced in HTML. - Modified date —
dateModified, the most-watched signal when a page is updated rather than replaced. - Visible date stamps — dates shown in the page body or byline, which crawlers and LLMs read as ground truth.
- Structural updates — new sections, new internal links, updated FAQ, changed headings. These move the page on its last-modified axis even when the body text barely shifts.
- Content refresh depth — how much of the substance actually changed. A bumped date with no body change is increasingly detectable by engines that compare versions.
- Surrounding signals — fresh inbound links, mentions across the AI answer evidence layer, recent press, recent reviews. Freshness is partly ambient.
A page can be technically fresh (modified date updated yesterday) but substantively stale (no real change). The reverse also happens: a page genuinely overhauled but with no date metadata exposed. Both fail. Freshness as a citation signal requires both the work and the visible proof of the work.
Why fake dates backfire
The temptation is obvious. Update the dateModified field across the site every month, change nothing else, watch the citations roll back in. It does not work — and the failure mode is worse than doing nothing.
Three reasons.
First, engines compare versions. When an engine has indexed the same URL multiple times, a date bump without a content delta is detectable. Some engines downgrade trust on URLs that show this pattern. Others ignore the date and fall back to inferred recency from links and mentions.
Second, the LLM reads the page body. If the body says “in 2022” and the metadata says “modified 2026”, the contradiction is visible. Models flag this as low-confidence and often drop the source from the answer.
Third, the cost is downstream. A page that lies about its freshness, once detected, becomes harder to rehabilitate than one that simply admits its publish date. Trust signals are sticky in both directions.
The honest version of the move is a content refresh — actual edits, new evidence, new examples — paired with an updated modified date. This is slower and more expensive than a metadata sweep. It is also the only version that holds up under retest, which is why Capston QA standards require freshness proofs to be auditable.
Building a refresh cadence
Not every page deserves the same cadence. A refresh strategy that touches everything dilutes signal and exhausts the content team. A strategy that touches the right pages, at the right interval, with real change, is where the gain sits.
Capston Core sequences refresh work in four tiers.
- High-citation, high-commercial-risk pages — the pages that AI engines already cite, where the brand has commercial exposure (booking, conversion, reputation). Refresh cadence: quarterly minimum, with substantive change.
- High-impression, low-citation pages — pages that surface in answers but are not cited. Often a freshness problem. Refresh cadence: every four to six months, focused on date metadata correctness and a real content delta.
- Authority backbone pages — methodology, certification, scoring. Refresh when the substance changes, not on a calendar. Document the change clearly.
- Tactical pages — campaign, seasonal, market-specific. Refresh tied to the underlying event; sunset honestly when the event is past.
The cadence is set per page, logged, and tied to a measurable outcome — a retest of the prompts where the page should be cited. Without the retest, refresh becomes ritual.
How this fits into Capston Core
Freshness is one signal among several, but it is one of the few that brands fully control. Capston Core treats freshness as part of the measurement loop: detected through the AI answer evidence layer, prioritised inside the Capston Core methodology, and audited under the Capston QA standards so that refresh work survives a retest rather than decorating it.
→ Back to Capston Core
FAQ
Does updating only the modified date help?
Marginally and briefly, then it backfires. Engines that compare versions detect the absence of a content delta and downgrade the URL. Pair every date change with a real edit.
How often should we refresh a positioning page?
For a high-citation, high-stakes page: quarterly with substantive change. For backbone pages: when the substance changes. A calendar without a content reason produces churn, not freshness.
Which engine cares most about freshness?
In the Chen et al. (2025) tests, Perplexity weighted freshness most heavily. ChatGPT and Gemini were more tolerant of older authoritative sources. Google AI Overviews fell in between.
Should we hide old publish dates?
No. Show both datePublished and dateModified honestly. Hiding the original date breaks trust signals and complicates audits. Engines increasingly cross-check dates against archived versions.
Reference
Chen, J., Wang, R., Chen, K., & Koudas, N. (2025). On the systematic evaluation of generative answer engines. arXiv:2509.08919v1. The study identifies freshness as one of four dimensions on which AI answer engines diverge significantly, alongside domain diversity, cross-language stability, and sensitivity to query phrasing.
Final CTA block
Find out if AI engines treat your pages as fresh.
Audit your freshness signal
Read the methodology