Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors

Content strategy · SEO and AI search

Your website has older articles, a limited maintenance budget and advice to “keep everything fresh”. Start with the pages where an outdated fact could change a customer’s decision—not the oldest date in your archive. Some pages need an urgent correction; others deserve to stay exactly as they are.

Three pages, one afternoon: what gets updated?

Consider a fictional consultancy with three pages waiting for attention. Its two-year-old explanation of a discovery workshop still describes the actual process accurately. A six-month-old software comparison recommends a plan whose export feature has since moved to a different tier. Last month’s service page promises a turnaround the consultancy can no longer meet.

The service promise comes first. Correct it before another prospect relies on it, even if that page attracts few visits. Next, repair the comparison: check the affected feature, revise the recommendation and explain who now needs a different option. Leave the workshop explanation alone unless a specific gap has emerged.

An oldest-first queue would reverse those priorities. A traffic-only queue could miss the misleading service promise. The useful order is decision risk first, relevant exposure second, effort third. This is an editorial prioritisation rule, not a search-engine scoring formula.

Magnifying glass highlights corrected claims on a comparison page while a timeless article and calendar remain aside
Refresh the evidence that changed; do not make the calendar the editor. Original illustration.

What the AI-citation evidence actually tells you

Ahrefs’ July 2025 citation study examined approximately 17 million cited URLs and search results. Its headline comparison reported an average cited-URL age of 1,064 days for the grouped AI assistants versus 1,432 days for organic search. That is a difference in observed age—not a 25.7% increase in the chance of getting cited.

The study used crawler first-seen dates as a publication-age proxy and database dates for updates. It did not randomly refresh equivalent pages and measure the causal effect. Platforms also differed, and the average cited page was still years old. Treat the findings as a reason to check relevance, not proof that weekly edits improve visibility. This is older research, not a new September 2026 platform announcement.

For a business website, the defensible aim is simpler: keep recommendations reliable when people encounter them through search or an AI answer. A citation to obsolete advice is not a win.

Refresh the recommendation, not just the sentence

Here is a fictional example of the software comparison’s repair. The plan names and conditions below are illustrative, not statements about a real product.

Before: “The Starter plan is the best choice for teams that need monthly exports.”

After checking the current documentation: “Starter no longer includes exports. If a monthly export is essential, compare the next tier’s total cost with a tool that includes it. If your team works entirely in the dashboard, Starter may still meet the requirement.”

The meaningful change is the decision boundary: which reader should choose which option, and why. In a real article, link to the current provider documentation, verify the relevant region and billing terms, and update any summary that still recommends the old choice. Correcting a paragraph while leaving the headline, comparison box or translated version unchanged creates conflicting advice.

Do not rewrite unaffected explanations to make the update look larger. Preserve useful examples and the existing URL when the page still answers the same question. If the question itself has changed, decide whether a separate article would serve readers better before merging unrelated intents.

Give each page a reason to re-enter the queue

Start a small maintenance record with the page URL, the claim most likely to become wrong, its authoritative source, the person responsible and the event that should trigger another review. For a product comparison, that event could be a pricing or feature change. For a service page, it could be a change in scope, availability or delivery time. For a conceptual explainer, it might be a recurring customer misunderstanding.

Review frequency follows volatility. A monthly source check may be a sensible starting point for a fast-changing tool comparison; a stable process explainer may only need a quarterly check and review when the process changes. Those are suggested operating cadences, not SEO requirements. Known misleading claims should not wait for the next scheduled review.

When no claim is wrong, prioritise a proposed improvement only if you can name the reader benefit. Does it answer an unresolved buying question? Replace an obsolete example? Explain an important limitation? If the only benefit is “a newer date”, leave the page out of the work queue.

Make the update date a promise you can defend

Google’s byline-date guidance recommends clear visible dates and consistency with publication and modification dates in structured data. It also explains that Google considers multiple signals and does not guarantee which date appears in search.

After a substantive revision, a short change note can help a reader understand what was checked. Preserve the original publication date where useful; label the update separately. Do not imply that every source and recommendation was revalidated if you only repaired one section.

Example maintenance note: “Updated the plan comparison and recommendation after checking export availability. The workshop example is unchanged.” A useful note describes the scope of the work, not a claim that the whole article is newly written.

Google’s people-first content guidance also challenges changing dates to make substantially unchanged pages appear fresh. The practical implication is to let the evidence determine the edit and let the edit determine the date.

Judge the refresh against the problem it fixed

Before editing, record the incorrect claim and the result you want: fewer enquiries based on an unavailable service, a comparison that reflects current terms, or a clearer next step for a specific buyer. Afterwards, confirm those conditions directly. When enough data accumulates, compare relevant enquiries and search performance over comparable periods, noting seasonality and other changes. A traffic rise alone does not prove the refresh caused it.

If you cannot tell which pages influence real enquiries, an SEO audit focused on page priorities can establish that baseline. The maintenance plan should then support your SEO and GEO strategy, not become a separate race to update everything. For pages that need a stronger reason to visit, see how to make service pages worth clicking after AI answers.

Questions about content refreshes

How often should website content be refreshed?

Use the rate of factual change and the consequences of being wrong, not a universal schedule. Review volatile prices and product capabilities when their sources change; revisit stable explanations when customer questions or the underlying evidence change.

Does changing the date improve AI citations?

The cited observational research does not establish that changing a date causes more citations. Update the substance when needed and keep publication and modification dates honest.

Should I create a new URL for an updated article?

Usually keep the existing URL when the reader question and purpose remain the same. A genuinely different question may need a separate page. Do not create a new annual copy merely to look recent.