"Does freshness matter for AI visibility?" is a fair question to be skeptical about. SEO spent a decade arguing about it. For AI answers the evidence is better than you might expect, but it is specific: freshness matters where answers are built by retrieval, and it matters in ways you can actually control.
What the evidence says
Two kinds of data point the same direction.
1. Reranking experiments. A 2025 study accepted at SIGIR-AP ("Do Large Language Models Favor Recent Content?") tested seven models, including GPT-4o and Llama 3, as passage rerankers. When passages of identical relevance carried visible dates, the models consistently promoted the newer passage; injecting dates flipped the preference between equally relevant passages by up to 25% on average. That is a controlled result, not a correlation: the models treat "newer" as a proxy for "better" when they can see it.
2. Citation-age distributions. Industry analyses of what AI engines actually cite (for example Seer Interactive's recency study) find the same skew in the wild: roughly two thirds of citation hits point at content published or updated within the last year, and about four in five within two years. Our own citation dataset shows the engines differ a lot in how they source answers, so treat exact percentages as directional, not universal.
Where to stay skeptical
Three honest caveats, because a number without its limits is marketing, not measurement.
- Citation-age data is partly correlational. Fresh pages are also more promoted, more linked and more relevant to news-flavored prompts. The reranker experiments isolate the bias; the distribution studies do not.
- The effect lives in retrieval, not in the model's memory. When ChatGPT search, Perplexity or Google AI Overviews build an answer, they fetch and rerank live pages, and that is where recency wins. A model answering purely from training data does not care that you updated a page last week.
- Industry matters. Evergreen reference topics age more gracefully than product comparisons or pricing pages. The skew is strongest where the question implies currency.
The part most sites get wrong
Engines can only see the recency you actually emit. A page that was genuinely rewritten last month but carries no dateModified, no article:modified_time and no visible date is indistinguishable from a page abandoned in 2022. In our audits this is one of the most common gaps: the work was done, the signal was never shipped.
The machine-readable signals that count:
dateModified(anddatePublished) in your Article or WebPage JSON-LDarticle:modified_timein Open Graph metadata- a visible
<time datetime>on the page, ideally with a short "what changed" note - an accurate
lastmodin your sitemap
A refresh playbook that is not date-bumping
Changing the date without changing the page is the cargo cult version of this. It does nothing for retrieval quality and it burns trust if anyone compares versions. A refresh that earns citations:
- Update the substance: data points, examples, screenshots, prices, names of current tools. If the facts did not change, say when they were last verified.
- State what changed near the top. Engines quote pages that make currency explicit; readers reward it too.
- Answer the core question in the first hundred words. Refreshes are the cheapest moment to fix a buried answer.
- Ship the signals: dateModified in schema, modified_time in OG, sitemap lastmod. Then resubmit the URL in Search Console.
- Start with the pages already winning citations. Defending a page AI already cites is far cheaper than earning a first citation.
How llemmy tracks this for you
Freshness is now measured in three places. The GEO audit checks whether a page emits machine-readable date signals at all and how old the latest one is. Content Studio has a freshness panel that ranks your own cited pages and WordPress posts by last update, flags anything untouched for a year, and hands you the refresh checklist above. And the WordPress plugin reads post modified dates directly and can emit the schema signals for you. None of this promises a ranking; it makes the one freshness lever you control visible and honest.