Content Freshness Signals

We're measuring whether your pages carry machine-readable dates that AI engines can parse and trust. Not "does a date appear on the page" -but "can a machine extract it?" Three distinct timestamp layers get examined: JSON-LD datePublished and dateModified within Article, BlogPosting, or WebPage schemas; HTML <time> elements with valid datetime attributes; and HTTP Last-Modified response headers.

The core metric is the "freshness coverage ratio" -the percentage of content pages (excluding static pages like contact and privacy) that include at least one machine-readable timestamp. A secondary metric, "freshness accuracy," checks whether the timestamps are plausible. A dateModified that precedes datePublished? Future publication dates? A lastmod untouched for 3+ years on clearly updated content? All count as inaccurate signals.

We also measure "timestamp consistency" across the three layers. When JSON-LD says the article was modified January 15 but the HTTP header says December 3, AI engines get conflicting signals. The consistency score tracks how often the sources agree (within a 24-hour tolerance).

This is critical for industries where information changes fast -healthcare, finance, tech, legal. Perplexity explicitly shows "last updated" in citations. Pages without dates get deprioritized or flagged with a warning that the information may be outdated.

AI answer engines use recency as a ranking signal when multiple sources answer the same query. For time-sensitive queries like "best live chat software 2026" or "current patient advocacy regulations," undated content gets systematically deprioritized. If your competitors publish dated articles updated within the last 90 days and your equivalent content has no dates at all, the AI picks them.

The penalty compounds at the domain level. When AI engines observe that a domain consistently lacks timestamps, they lower the domain's overall freshness trust score. Even your genuinely current content gets disadvantaged because the domain pattern suggests poor date hygiene.

Freshness signals hit Google AI Overviews directly. Google's systems use dateModified to decide whether to include a source. Pages with recent modification dates are weighted higher for queries where recency matters. A site where 80% of pages carry no modification date is telling Google: "I can't confirm when any of this was last verified."

There's a deceptive practice risk too: sites that update dateModified without making real content changes are "timestamp farming." We cross-reference dateModified against actual content changes between crawl snapshots to catch this. Honest freshness signals -where the date reflects meaningful updates -build long-term trust.

Every crawled page gets classified into one of three categories: content pages (articles, blog posts, guides, product pages), structural pages (category listings, tag archives, search results), and static pages (about, contact, privacy, terms). Only content pages are included in the freshness calculation -static pages aren't expected to carry publication dates.

For each content page, timestamps get extracted from all available sources. JSON-LD properties from Article, BlogPosting, NewsArticle, and WebPage schemas -specifically datePublished, dateModified, and dateCreated. HTML <time> elements with their datetime attributes. The HTTP Last-Modified header from the server response. Meta tags like <meta property="article:published_time"> and <meta property="article:modified_time">.

Each timestamp is validated for format (ISO 8601), plausibility (not in the future, not before the domain's registration date), and cross-source consistency. We flag specific issues: missing dateModified when datePublished is present (suggests the content was never updated), dateModified identical to datePublished across all pages (that's a template default, not actual tracking), and timestamps differing by more than 24 hours across sources.

The results include a histogram showing the age distribution of your content -how many pages were last modified within 30 days, 90 days, 6 months, 1 year, and beyond. This reveals whether your site maintains active content or has a massive "stale tail" of aging pages that haven't been touched.

Freshness scoring combines three sub-metrics:

1. Freshness coverage (40% of score): - 90-100% of content pages have at least one valid timestamp: 4/4 points - 70-89%: 3/4 points - 50-69%: 2/4 points - 30-49%: 1/4 points - Below 30%: 0/4 points

2. Timestamp accuracy and plausibility (30% of score): - All timestamps pass validation: 3/3 points - Less than 10% have issues (future dates, impossible sequences): 2/3 points - 10-25% have issues: 1/3 points - More than 25% have issues: 0/3 points

3. Cross-source consistency (30% of score): - JSON-LD, HTML, and HTTP timestamps agree within 24 hours on 90%+ pages: 3/3 points - Agreement on 70-89%: 2/3 points - Agreement on 50-69%: 1/3 points - Below 50% agreement or only one timestamp source present: 0/3 points

Deductions: - -1 point if more than 50% of pages have dateModified identical to datePublished (no real update tracking) - -1 point if no pages use the HTML <time> element (missing the visible date layer)

Total normalized to 0-10. Sites relying exclusively on CMS-generated dates without structured data integration typically land between 2 and 4.

Sites scoring 8+ on content freshness have a systematic approach: their CMS automatically generates datePublished and dateModified in Article JSON-LD, their templates include HTML <time> elements with datetime attributes, and their server returns accurate Last-Modified headers. These three layers align within 24 hours on 90%+ of pages. The result is a domain that AI engines trust for recency-sensitive queries.

Sites scoring 2-3 typically fall into one pattern: dates exist only as visible text ("Published January 2026") without any machine-readable equivalent. A human can read that date, but ChatGPT's retrieval system, Perplexity's indexer, and Google's AI Overview pipeline cannot extract it reliably. The page is effectively undated from an AI perspective.

The difference between a 5 and an 8 often comes down to dateModified tracking. Many CMS platforms set dateModified equal to datePublished by default and never update it. A site with 200 articles all showing the same dateModified as datePublished tells AI engines that no content has ever been refreshed - even if the author has made substantial updates. Enabling real modification tracking in the CMS is typically a one-time configuration change worth 2-3 points.

Timestamp farming is the most penalized mistake. Some sites update dateModified daily across all pages without making actual content changes, hoping to appear fresh. AI engines detect this pattern - when every page on a domain shows yesterday's date but crawl comparisons show zero content changes, the freshness signal is discredited. The site may score worse than one with no dateModified at all.

Missing HTML <time> elements cost a consistent 1 point. Many sites include dates in JSON-LD but display them as plain text ("Last updated: Feb 10, 2026") without wrapping them in <time datetime="2026-02-10">. The visible date helps humans but gives AI crawlers that parse rendered HTML no structured timestamp to extract.

Future dates in datePublished are surprisingly common, especially on sites using timezone-unaware date generation. A server in UTC generating a date for a US-based publisher may produce tomorrow's date. AI engines flag future publication dates as implausible and discount the entire page's timestamp reliability.

Conflicting timestamps across layers destroy the consistency sub-score. JSON-LD says January 15, the HTTP Last-Modified header says December 3, and the visible text says "Updated February 1." Three dates, three stories. AI engines cannot determine which is accurate and default to treating the page as having no reliable timestamp.

Perplexity displays "last updated" timestamps prominently in its citations. When a source has a clear, recent dateModified, Perplexity shows it alongside the citation - building user trust and increasing click-through. Sources without timestamps get a generic display that communicates less authority. For time-sensitive queries, Perplexity explicitly filters for sources updated within the relevant timeframe.

ChatGPT's retrieval pipeline uses freshness as a ranking signal when multiple sources provide equivalent information. For a query like "best project management tools 2026," a page with dateModified in the last 30 days ranks higher than an identically relevant page with dateModified from 18 months ago. The fresher page is assumed to reflect current information.

Google AI Overviews use dateModified extensively for determining which source version to feature. When the same topic is covered by multiple domains, the AI Overview preferentially cites the most recently modified source - provided the modification appears genuine (content hash changed, not just a timestamp bump). This makes honest freshness tracking a competitive advantage in AI Overview placement.

Claude evaluates timestamp presence as a credibility indicator. Content with complete date metadata (both published and modified dates in structured format) receives higher trust scores in retrieval ranking than undated content, especially for queries where temporal context matters - industry trends, regulatory changes, technology comparisons.