AI Engines Prefer Fresh Content. Local Business Pages Are the Stalest on the Web.

Ahrefs published one of the clearest empirical looks at recency behavior in AI search in July 2025: 16.975 million cited URLs across ChatGPT, Perplexity, Gemini, Copilot, Google AI Overviews, and traditional organic SERPs. The topline is neither hype nor nothing. Freshness bias appears real, but moderate.

On average, AI-cited pages were 1,064 days oldAhrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 at publication date versus1,432 daysAhrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 for organic SERPs, a 25.7%Ahrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 freshness advantage. By last-updated date, AI-cited pages averaged 909 daysAhrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 versus 1,047 daysAhrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 for organic, or 13.1%Ahrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 more recently updated.

Those percentages are meaningful, but they are easy to overstate. The typical AI-cited page is still about 2.9 yearsAhrefs, AI Assistants Prefer to Cite Fresher Content, Jul 2025 old. This is not a world where every assistant only cites yesterday's article.

The sharper strategic implication is not for media publishers with content teams. It is for local and SMB operators whose service and location pages often sit untouched for years. If assistants that mediate discovery are even modestly freshness-sensitive, the least frequently updated businesses are structurally exposed.

The platform split matters more than the aggregate. Ahrefs reports a broad mean effect, but engine-level behavior ranges from strongly fresher than organic to slightly older than organic.

The ordering behavior is also non-trivial. Ahrefs observed that Perplexity and ChatGPT tend to order in-text references from newer to older, while organic results and some other surfaces often place older content earlier. This suggests freshness is not only affecting whether a page is cited, but also where it appears in the support stack a user is most likely to scan.

This section is interpretation, not an Ahrefs claim. Ahrefs establishes a measurable freshness tilt. We are extending that signal to a business type that is operationally weak on content maintenance: local and small-service organizations.

Many local sites are launched once, then largely frozen. Core service pages remain unchanged for long periods. Location pages are cloned and left untouched. Teams often have no in-house content operator, no editor, and no recurring update process tied to real-world service changes.

In classic SEO, those pages could still perform if backlinks and local pack factors carried enough weight. In assistant-mediated discovery, that static posture is riskier. If model retrieval and citation layers weight recency even moderately, stale pages lose selection probability at the margin, especially against regional publishers or aggregators publishing continuously.

This is exactly the kind of question current public studies cannot quantify rigorously: not just whether AI likes fresher content overall, but whether recency effects are larger in specific industries, city tiers, or intent classes such as “near me” service decisions.

The mean cited page at 1,064 days old should reset expectations. Recency helps, but AI systems are not replacing depth and authority with hourly updates.

Google AI Overviews top-3 averaged 1,432 days since publication, slightly older than organic in Ahrefs' table. Given Google still mediates most user demand, many businesses should avoid over-rotating into high-frequency update routines that weaken page quality.

John Mueller has repeatedly cautioned against changing publish/update dates without substantive content changes. Date-bumping is not a strategy; it is a trust and quality signal risk.

Ahrefs itself notes that publishing genuinely useful new pages often outperforms endlessly tweaking older pages. Recency should be integrated with quality, topical coverage, and intent match, not treated as an isolated optimization target.

If you run a local or SMB content program with limited capacity, the answer is not “update everything monthly.” It is an evidence-based cadence tied to commercial pages and recurring question clusters.

This is an operational discipline problem, not a writing-style problem. Teams that build a maintenance cadence around real-world business change should accrue recency benefit without sacrificing quality. Teams that chase freshness theatre will spend budget and gain little.

A practical way to keep this honest is to treat every update as a hypothesis with a tracked outcome. If a service page is revised, define the expected impact first: better answer eligibility for a specific question class, cleaner citation context for specific facts, or reduced mismatch between offer language and user prompts. Then measure whether those outcomes appear over a fixed window. Without that loop, “freshness work” becomes routine copy edits disconnected from discovery performance.

Local teams can also reduce maintenance burden by structuring pages for additive updates instead of rewrites: stable core sections, timestamped change blocks for material updates, and explicit service-difference notes by location when appropriate. This keeps content truthful and maintainable while still signaling that the page reflects current business reality. The goal is not velocity for its own sake; it is durable accuracy that can be retrieved and cited confidently.

The larger unresolved question is geographic and longitudinal: where freshness sensitivity is strongest, for which query intents, on which engines, and whether that sensitivity rises over time as assistants mediate more discovery. Answering it requires measuring these effects directly, by place and by question, instead of relying on static averages.