Is it enough to simply change the date?

No. A new date without a real revision is detected by search engines and AI systems and can actually harm you. Freshness must be backed by genuine content maintenance: corrected facts, new examples, current figures. Only then do you deliberately set the update date anew.

Do all contents have to be constantly updated?

No, that depends on the topic. Time-critical content like prices, deadlines or product versions needs frequent maintenance. Stable foundational topics (so-called evergreen content) require a less frequent review. Prioritize the pages for which you want to be cited in AI search.

Freshness

Freshness is how current and actively maintained a page appears to be: when it was published, when it was last genuinely revised, and whether the facts on it still hold. AI assistants and AI Overviews weigh this heavily for anything time-sensitive - prices, deadlines, product versions, legal thresholds - because a wrong answer built on stale content is worse than no answer at all. Freshness alone doesn't earn you a citation, but on a fast-moving topic, being visibly out of date can get you dropped from consideration entirely.

Why freshness matters for AI visibility

Systems like ChatGPT, Perplexity, Gemini and Google's AI Overviews are built to answer confidently, and confidence on a changing topic requires current information. Google's AI Overviews alone reach over 2 billion monthly users across 200+ countries, and ChatGPT has crossed 900 million weekly users - at that scale, an engine that keeps citing outdated prices or superseded rules erodes its own credibility fast, so the underlying models are tuned to prefer sources that look actively kept up. That preference cuts against you when a page hasn't been touched: a model that notices a stale date, a superseded figure, or a fact that's been publicly corrected elsewhere will route around you toward whoever updated first. Freshness is one of several trust signals feeding into whether you get selected as a source - not a ranking trick, just table stakes for staying eligible on anything that changes.

How freshness is detected technically

Crawlers and models read a mix of visible and structural signals: the publication and update date on the page itself, dateModified and datePublished in structured data where present, how often a crawler like GPTBot or Googlebot finds something changed, and content-level cues such as references to the current year or recent events. It's worth being precise here: Google has stated plainly that no special markup, schema, llms.txt file, or AI-specific content is required for AI Overviews or AI Mode, and it explicitly warns against writing separate content "for AI" rather than for readers. Structured dates help a machine parse what you already did; they don't substitute for actually having done it. A changed timestamp with unchanged substance is easy to spot and doesn't earn trust - what actually registers is real editorial work: corrected numbers, added detail, removed claims that no longer hold.

Common mistakes

The most common mistake is cosmetic re-dating - bumping the visible date or dateModified field without changing anything underneath. It's transparent to anyone checking, human or model, and it burns trust once caught. The opposite failure is just as common: publishing a solid guide once and never revisiting it, which quietly ages out pricing pages, comparisons, and how-to content built on version numbers. A related error is chasing freshness through volume - shipping thin new posts on a schedule instead of maintaining the handful of pages that actually matter for citation. Some teams also reach for schema markup or an llms.txt file as a freshness shortcut; neither is a recognized ranking or citation factor, and an Ahrefs analysis of roughly 137,000 sites publishing llms.txt found about 97% saw no measurable referral traffic tied to it. And plenty of pages simply contradict themselves - visible text claiming one year, schema metadata claiming another - which is its own small credibility problem.

Maintaining freshness systematically

Treat freshness as a maintenance cadence, not a one-time task. Keep a short review calendar tied to how time-sensitive each page is: quarterly checks for pricing, regulatory, or version-specific content; annual checks for stable, evergreen material. Each real revision should touch the substance first - verify figures, update examples, remove anything superseded - and only then update the visible date and dateModified together, so the two never disagree. Spend that maintenance effort on the pages you actually want cited, not everything at once; a small set of well-kept core pages will outperform a large pile of neglected ones. The goal is ordinary: genuine upkeep, reflected honestly in the metadata, applied to the pages where being current actually changes whether you get picked as a source.

Example

A small tax advisory firm in Leipzig publishes a guide to filing deadlines and flat-rate allowances. For a year it ranks well and gets cited by Perplexity. Then the thresholds change nationally, but the firm's page stays untouched. The AI assistant starts naming a competitor instead - one whose article carries the new figures and a recent update date. The firm goes back in, corrects the amounts, adds a line noting the legal change, and updates dateModified to match. Within a few weeks it's being cited again. The underlying guide was never bad; it just stopped being current, and that was enough to lose the spot.

Common questions

Is changing the date enough on its own?

No. A bumped date with no real revision behind it is easy to detect and can hurt more than help. Freshness needs to be backed by an actual edit - a corrected fact, a new figure, an updated example - and only then should the date reflect it.

Does every page need constant updates?

No. It depends on how time-sensitive the topic is. Prices, deadlines, and product versions need frequent attention; stable evergreen explainers need far less. Focus your review effort on the pages you most want an AI assistant to cite.