How freshness signals shape AI answers

A freshness signal is a piece of evidence that tells an AI engine how recently a page's content was created or updated: a visible publish date, a dateModified field in structured data, a sitemap lastmod timestamp, or a changelog that shows the page is still maintained. These signals matter because AI assistants demonstrably prefer recent content, and the preference is strong enough to decide which sources get pulled into an answer before the model ever weighs what they say.

The cleanest evidence for that preference comes from Ahrefs, which analyzed roughly 16.975 million cited URLs across ChatGPT, Perplexity, Gemini, Copilot, AI Overviews, and Google's organic results. AI-cited pages averaged 1,064 days old against 1,432 days for the URLs ranking in Google's top-10 organic results, which makes AI citations about 25.7% fresher, roughly a full year newer on average. The gap was widest for ChatGPT, which cited content that was 458 days newer than typical organic results.

#Why recency acts as a filter, not a tiebreaker

It is tempting to read freshness as a small bonus, the thing that breaks a tie between two otherwise equal pages. The retrieval data suggests it works earlier than that. When several sources cover the same question, the system tends to pull the newer ones into the candidate set first, so stale pages get dropped before their substance is evaluated at all.

A controlled study makes the mechanism explicit. In "Do Large Language Models Favor Recent Content?", Fang and colleagues tested seven models (GPT-3.5-turbo, GPT-4o, GPT-4, LLaMA-3 8B and 70B, and Qwen-2.5 7B and 72B) on a reranking task and found that injecting fresh publication dates shifted the mean publication year of the top-10 results forward by up to 4.78 years. Larger models showed a smaller effect, but none eliminated it. The date itself, independent of the text, moved the ranking. That is the behavior of a filter, not a tiebreaker.

Field data lines up with the lab result. Seer Interactive's analysis of AI crawler logs found that nearly 65% of bot hits landed on content from the past year, 79% on content from the last two years, and 89% on content updated within the last three. If your page sits outside that window, you are competing for a shrinking slice of attention.

#The engines do not weight freshness equally

Recency is not one setting applied across every assistant, so a single freshness strategy will land differently depending on where you want to be cited. Seer's study found that about 50% of Perplexity's citations came from 2025 content alone, while ChatGPT spread its citations wider, with roughly 31% from 2025 and a long tail reaching back to much older reference pages. Google's AI Overviews sat in between, with about 44% of citations from 2025.

The practical reading: Perplexity rewards a page that was genuinely updated in the last quarter more than ChatGPT does, while ChatGPT will still cite a durable, source-backed page that has been stable for years. If Perplexity visibility is your goal, a real update cadence matters more. If you care about ChatGPT, depth and sourcing carry more of the load and freshness is a supporting signal rather than the deciding one.

#What a real freshness signal looks like

The signal engines respond to is a coherent set of timestamps that all agree the page actually changed. That means four things moving together: a dateModified value in your JSON-LD that reflects a real edit, a matching lastmod in your XML sitemap, a visible "updated on" date in the page itself, and content that actually differs from the prior version. When those agree, you are giving the crawler consistent evidence. When they disagree, for example a dateModified of last week sitting on a page whose body has not changed since 2023, you are giving it a reason to distrust the date.

This is the part teams get wrong. Bumping dateModified on a schedule, with no change to the underlying content, is not a freshness strategy. It is a date that contradicts the page. In our audits the pages that hold citations over time are the ones where the timestamp tracks a real edit: a new data point added, a figure corrected, a section rewritten because the product or the market moved. The date is a claim about the content, and the content has to back it up.

Changelog and "what's new" pages earn their citations the same way. A changelog is valuable to an AI engine precisely because it is a structured, dated record of real changes, which is the strongest possible freshness evidence for a product. It also gives the crawler a frequently-updated URL to return to, which keeps the rest of your documentation in the recrawl rotation.

#How to put this to work

Freshness is not a one-time fix, it is a cadence. Identify the pages where recency actually matters, which are the ones answering questions whose correct answer changes over time: pricing, comparisons, integrations, version-specific guidance, anything tied to a fast-moving market. Put those on a real review schedule. When you revise them, change the content first and update dateModified, the sitemap lastmod, and the visible date to match. Leave genuinely evergreen explainers alone rather than faking edits on them.

Then measure whether it worked. Freshness is an input; citation share is the outcome you actually care about. If you are not already tracking how often AI answers cite your pages, our guide to measuring GEO with citation share covers how to sample for it, and the structured data AI engines actually read explains where dateModified fits among the schema fields that matter. Update with intent, keep the timestamps honest, and let the trend line tell you whether the engines noticed.