Recent beats polished
Why recency dominates trust
An AI assistant that cites an out-of-date price, a closed business, or a discontinued product gets corrected by users and quietly down-weights the source. Over many such corrections, the assistant learns to favour pages that prove they are current. Recency, in this context, is not vanity. It is the ground truth that lets an assistant answer time-sensitive queries — pricing, availability, opening hours, regulatory rules — without producing a confidently wrong answer. A small page updated last month often outranks a thorough page last touched in 2022.
The visible signals: dates on the page
Visible last-updated dates near the top of articles, product pages, and reference content are read by both humans and machines. They give AI systems a direct integer — number of days since modification — that is easy to compare across candidate pages. The convention is straightforward: a 'Last updated:' line in the byline area, paired with a 'Published:' date if the two differ. Faking the dates by changing them without updating the content is, predictably, a flag that AI systems and search engines now detect and penalise. Honest updates beat decorative ones.
The machine-readable signals: dateModified, lastmod, headers
Three machine-readable signals reinforce visible dates. JSON-LD's dateModified property on Article and Product schema. The lastmod element in sitemap.xml entries. The Last-Modified HTTP header returned with each page response. Vercel's agent-readability framework treats all three as positive signals. Most content-management systems update them automatically, but it is worth verifying that they actually fire — sites with a stale 1970-01-01 sitemap or a missing dateModified are common enough to be a frequent cause of unexplained visibility loss.
A sustainable maintenance routine
Trying to update everything continuously is unsustainable for most sites. A workable rhythm is a quarterly review of high-value pages — pricing, key product pages, FAQs, the home page — and an annual review of everything else. The cycle has three steps. First, check the facts: prices, dates, regulatory references, product names, broken links. Second, refresh the dates honestly when changes are made. Third, log what was reviewed even when nothing changed; the log demonstrates due diligence and is itself a form of freshness signal when published as a 'last reviewed' line. The cost is one or two hours per page per year, recovered many times over in ongoing visibility.
When was the last time you updated the most-trafficked page on your site? If the answer is more than six months ago, an AI assistant is probably already discounting it.