How much fresher AI citations are
AI systems cite markedly newer content than Google for the same category. Researchers at the University of Toronto extracted publication dates from every source cited across 100 ranking queries per vertical, in a study presented at the EDBT/ICDT 2026 workshops, and compared the median age.
| Vertical | Claude 4.5 Sonnet | GPT-4o | Perplexity Sonar Pro | |
|---|---|---|---|---|
| Consumer electronics | 62 days | 80 days | 90 days | 130 days |
| Automotive | 148 days | 162 days | 217 days | 493 days |
The automotive gap is the telling one. Google leans on content averaging 17 months old; Claude leans on content five months old. The researchers preferred explicit publication dates over modification dates when both were present, so these figures are conservative. AI systems also extracted dates from more of their sources (Claude 92.5% of cited URLs against Google's 61.5%), which suggests recency is an explicit input, not a side effect.
Why AI systems weight recency
AI systems weight recency because retrieval exists to fill the gaps in their training data. A model with a training cutoff supplements itself with live pages precisely to be current; an old page defeats the purpose. For ranking and recommendation queries, older content is also simply less accurate: discontinued products, outdated prices, missed developments.
Some platforms make this explicit. Iyappan's comparative analysis of SEO, AEO and GEO records recency as a "very high" weighted factor in Perplexity's profile, the highest weighting of any single signal for that platform. For businesses whose buyers research on Perplexity, freshness is a first-order signal.
The freshness window by sector
The practical freshness window is roughly 60–100 days for fast-moving categories and 150–220 days for slower ones. The Toronto medians give the anchors: consumer electronics at 62–90 days for AI systems, automotive at 148–217.
For professional services, software, AI-adjacent and financial categories, treat 130 days as the outer edge and 180 as clearly disadvantaged. For industrial, regulated and infrastructure categories, 250 days is the edge and 350 the limit. Google's tolerance (130–493 days) is why this loss is invisible in normal SEO reporting: the page keeps its ranking while it quietly drops out of AI answers.
Machine-readable dates are part of the window. If a page is genuinely current but carries no dateModified in its JSON-LD, no visible "last updated" line and no <time> element, the assistant may not credit it. The refresh has to be both real and declared.
A quarterly refresh routine
A quarterly review cycle keeps priority content inside the AI window without abandoning evergreen strategy. Evergreen pages still build topical authority and Google rankings; they just need a maintenance layer.
- Pick the priority set. Service pages, category explainers, FAQ pages and the educational pieces that target queries AI actually answers. Ten to twenty pages for most businesses.
- Update substantively. Current-year statistics with sources, new research, FAQ answers revised to the questions buyers ask now, service descriptions matched to what you deliver today.
- Declare the update.
dateModifiedin the Article schema, a visible updated date,<time>tags. Haddad's structured-content study gives revision visibility its own weight (0.10) in the completeness index that predicted AI inclusion. - Decide refresh versus new. When a page is far outside the window and no longer ranking, a new piece on the same topic is often faster than reviving stale date signals.
Freshness is an amplifier, not a substitute. Kargaev's ranking-factor analysis puts entity mentions, statistics and citations at the top of the GEO hierarchy; a new but vague page still loses. A refresh that adds evidence and updates the date compounds both. The content quality piece covers the evidence side; seasonal SEO covers timing-sensitive refreshes.
Editorial coverage ages the same way. A trade-press mention from eighteen months ago is outside the window for fast categories, which is one more reason to run digital PR as a continuous quarterly programme rather than a one-off campaign.
Frequently asked questions
How old is too old for AI citation?
Beyond about 130 days in fast-moving categories and about 250 in slow ones, a page is at the edge of what AI systems typically cite; past 180 and 350 days respectively it is clearly disadvantaged against fresher equivalents. These are medians, not cut-offs: strong content still gets cited later, but with falling probability.
Does updating a page reset its freshness?
Yes, if the update is substantive and declared. A cosmetic edit without a new dateModified and visible updated date may not register. Real changes plus updated machine-readable dates are what AI systems extract.
Which platform cares most about freshness?
Perplexity, which weights recency "very high" in its ranking profile and is built as a real-time research tool. Claude and GPT-4o also cite far newer content than Google. Gemini inherits some of Google's tolerance for older pages through Search grounding, but still skews newer than Google itself.
Does freshness matter equally in every industry?
No. The faster a category changes, the narrower the window: 62 days for consumer electronics against 148 for automotive in the Toronto data. Technology and professional services sit at the fast end; stable, regulated services at the slow end.
Should I delete content outside the window?
Rarely. Old pages that still rank in Google keep earning there. Refresh them if they matter commercially; publish new content if they have lost both freshness and rankings; delete only if the page is actively inaccurate.
Does this apply to evergreen explainers too?
Less. The finding is strongest for ranking, recommendation and "best of" queries, where currency affects accuracy. A definition of entity SEO does not go stale at the same rate. Prioritise commercial and comparison pages for the quarterly cycle.
Check how much of your content is still in the window
The free analysis reads your key pages and shows whether AI assistants currently cite you for your category, so you can see where a refresh would put you back in the answer.
