Articles ·

Content freshness for AI search — the 13-week rule and the 67% freshness lift

The one-line version: Roughly half of AI citations point to content published or substantially updated in the last 13 weeks, and recently updated pages earn 6 AI citations on average vs 3.6 for outdated pages — a 67% lift attributable entirely to freshness rather than new content. For Perplexity specifically, temporal freshness accounts for ~44.2% of selection weighting — higher than any other AI search engine measured. A 30-45 day refresh cadence keeps evergreen commercial content in the active pool.

The 13-week rule and how it was measured

The “13-week rule” — that roughly half of AI citations across engines point to content published or substantially updated within the last quarter — comes from Amsive’s citation research, which analyzed citation patterns across ChatGPT, Perplexity, Claude, and Gemini.

Two subsequent studies corroborated with larger samples:

Study Sample size Finding
Ahrefs, July 2025 17 million AI citations Cited pages averaged 25.7% fresher than pages ranking organically for the same queries
SE Ranking, 2026 216,524 pages ~44.2% of Perplexity’s selection weighting attributable to temporal freshness — highest of any AI engine measured
SE Ranking, 2026 Same sample ~half of Perplexity’s citations resolve to current-year content

The evidence converges: content freshness is a first-class ranking signal for AI citation, not a tiebreaker.

The refresh cadence table

From CiteFlow’s freshness playbook, matched to real content prioritization:

Content type Recommended cadence
Top 20% by citation potential (high-value commercial pages, pillar content) Every 30 days
Mid-tier pages Every 60-90 days
Long-tail pages Every 6 months, or when substantive change is warranted
Time-sensitive pages (pricing, comparisons, “best X in [year]”) Review monthly; refresh immediately when competitors move
Evergreen commercial content generally Every 30-45 days

Why 30-45 days for evergreen? Two reasons:

  1. Keeps dateModified inside the 13-week (94-day) active-freshness window
  2. Provides two full re-index cycles of buffer before falling out of the pool

The 67% citation lift figure — pages averaging 6 AI citations vs 3.6 for stale ones — is attributed specifically to freshness rather than new content creation. That’s the ROI on refresh work vs new content production.

What the breaking-topic window looks like

For time-sensitive queries (news, product launches, regulatory changes, model releases), the effective citation freshness window compresses dramatically. From CiteFlow’s log-level testing:

  • Breaking-topic content has a 48-72 hour half-life
  • A page published on Monday can have its citation share reduced by Thursday for the same query
  • Even if your content is objectively better, newer coverage will erode citation share within days

This is why reactive newsjacking (see digital PR for GEO) works: the window is so tight that speed dominates content quality. On breaking stories, be first with a credible take rather than definitive with a slow one.

Crawl and indexing timelines

Freshness only matters if crawlers can pick up the change. CiteFlow’s benchmarks:

Scenario Typical time
New domain → first citation eligibility 7-14 days
Established domain → PerplexityBot re-crawl 24-72 hours
Well-structured content on authoritative domain → live Perplexity response 24-48 hours
Established domain, clean rendering, good schema → citation appearance 24-48 hours
New domain → Perplexity eligible pool entry 7-14 days

New domains take longer than 7-14 days when they lack:

  • A Wikipedia entity (see entity-first SEO)
  • Mentions in the training corpus
  • Inbound links from already-crawled sources

Slow-discovery causes:

  • robots.txt blocks on PerplexityBot (see AI crawler decisions)
  • Aggressive edge caching serving stale HTML to bots
  • Client-side rendering that hides updates from crawlers

The freshness signal hierarchy

Not all timestamp signals carry equal weight. CiteFlow ranks Perplexity’s freshness signals by influence approximately as follows:

  1. <meta property="article:modified_time"> — most reliable machine-readable signal
  2. <meta property="article:published_time"> — second most reliable
  3. Schema.org dateModified and datePublished on Article, NewsArticle, BlogPosting
  4. Visible “Last updated” text near the title or byline — human-readable dates get parsed and appear to carry independent weight beyond schema
  5. XML sitemap <lastmod> — used more for crawl prioritization than direct ranking; helps trigger the re-crawl that lets other signals matter
  6. HTTP Last-Modified header — mostly relevant for conditional GETs

The point of the hierarchy: stack the signals. Consistent values across schema, visible dates, sitemap <lastmod>, and HTTP Last-Modified reduce ambiguity and reinforce each other. Mismatches between schema and visible dates get discounted.

The abuse patterns that suppress trust

CiteFlow specifically warns that:

  • Changing dateModified without changing content does not improve citations
  • Perplexity uses diff-aware crawling — cosmetic timestamp changes are ignored
  • Repeated abuse can suppress a domain’s freshness score
  • Backdating updates without editing is detectable and suppresses trust

A refresh should include at least one substantive change:

  • A new statistic
  • A rewritten paragraph of analysis
  • A new example
  • A verified claim that was previously unverified

If you’re materially rewriting a piece, update both datePublished and dateModified, or use an updated-date-forward display pattern (show “Updated Sep 18, 2026” prominently, mention original publish date lower down).

What each high-value refresh should include

For the top 20% by citation potential, the CiteFlow refresh checklist:

  • Update every statistic — verify against current primary sources, replace stale ones
  • Add one new subsection — a new angle, an emerging tactic, a competitive comparison
  • Verify every claim — check each cited source is still live and still supports the specific wording
  • Refresh the “date” language — “in 2025” → “in 2026”, “recent research” → “the [Month] 2026 study”
  • Update comparisons — if you mention competitors or alternatives, verify current pricing/features
  • Check for broken links — every 404 in a cited-source list reduces trust
  • Sitemap <lastmod> bump — combined with fresh sitemap submission noticeably reduces re-crawl latency

The Perplexity-specific advantage

Perplexity is described as the AI search engine with the strongest real-time bias. Per CiteFlow:

  • Perplexity issues live web queries against an index of 200+ billion URLs on nearly every prompt
  • Freshness is a first-class ranking signal, not a tiebreaker
  • ~44.2% of selection weighting is temporal freshness (SE Ranking 2026 study)
  • ~half of citations resolve to current-year content

Compare with the other engines:

Engine Freshness weighting Retrieval approach
Perplexity ~44.2% (highest measured) Live web query on nearly every prompt
ChatGPT Lower than Perplexity Mixed: training corpus + structured retrieval
Gemini Lower than Perplexity Index-heavy; blends AI Overview signals

For a Perplexity-focused GEO strategy, freshness is the single highest-leverage signal to optimize.

The sitemap resubmission play

The specific mechanic that reduces re-crawl latency after a content update:

  1. Refresh the content (substantive change, not cosmetic)
  2. Update dateModified in schema and visible byline
  3. Bump <lastmod> in the XML sitemap
  4. Resubmit the sitemap in Google Search Console and Bing Webmaster Tools (verification guide)
  5. Trigger IndexNow submission for that URL

Steps 3-5 are what convert “I updated the content” into “crawlers know to look again.” Without them, PerplexityBot might not revisit for 24-72 hours (established domain) or up to 14 days (new domain).

For sites at scale, this is worth automating. A simple build hook can bump <lastmod> for changed pages, resubmit the sitemap, and fire IndexNow on every deploy. The 24-48 hour crawl → 11-day citation window in MaxAEO’s unblocking data suggests the recovery-and-appearance loop closes fast when signals are stacked.

The refresh vs new-content tradeoff

The 67% citation lift from freshness (6 citations vs 3.6) suggests refresh work has better ROI than new content production on a per-hour basis, especially for the top 20% of pages by citation potential. But refresh doesn’t compound the way new content does — each refresh resets the 13-week clock without adding surface area.

A balanced allocation:

  • 60% of content hours on refresh for the top 20% of pages (30-day cadence on the highest-value 5, 60-day on the next 15)
  • 40% of content hours on new production — but only for content that fills genuine gaps or targets new query patterns

The 40% new-production allocation is where digital PR ties in: original research is the highest-ROI new content because it doubles as PR asset and citable statistical source.

The strategic point

Freshness is the highest-leverage signal in Perplexity’s algorithm (44.2% weighting) and a top-3 signal across all major AI engines. Yet most content sites treat freshness as a “publish and forget” activity — writing new pieces but not systematically refreshing top performers.

Set a 30-day refresh cadence on your top 5 pages. Set 60-day cadence on the next 15. Automate sitemap <lastmod> bumps and IndexNow submission on deploy. In 90 days you’ll have refreshed 20 pages twice each, with signals stacked across schema, visible dates, and sitemap — every one of those pages held inside the 13-week active pool.

Related reading: Perplexity source selection — the retrieval-and-rerank system that freshness feeds. Digital PR for GEO — the 25% earned-media citation surface where reactive newsjacking exploits the 48-72 hour breaking window. AI crawler decisions — the robots.txt configuration that determines whether crawlers can pick up your refresh at all. Search Console verification — the sitemap resubmission step that turns a content update into a re-crawl signal.