How to Manage Content Freshness for AI Search and GEO Teams

What does content freshness mean in AI search?

Content freshness in AI search means that a page's claims are still accurate, traceable and supported by evidence that machines can reach. It is an evidence-integrity standard, not a label that can be created by changing the date at the top of an article.

ChatGPT Search can search the web when a question would benefit from current information and can display links to the sources used. Perplexity similarly describes its product as a real-time web search system that answers with citations to original material. Both behaviours make one operational question more useful than “How many new articles did we publish?”: which existing pages remain dependable enough to support an answer?

A two-year-old definition can still be sound when its facts have not changed. A page published last week can already be stale if it lists a former executive, links to a withdrawn study or describes a service that is no longer available. GEO-targeted PR therefore depends on the quality of the relationship between an organisation, a subject and corroborating sources, rather than recency theatre.

FL PR & Communications applies the same logic to media authority. Its first-party content and professional-channel commentary frame AI visibility as a question of where a brand is mentioned, which sources support the mention and whether the subject association is correct. Freshness protects that association from quietly decaying.

Which changes justify a new modified date?

A new modified date is justified when the main claim, supporting evidence, decision guidance or structured representation of the page has materially changed. Google's sitemap documentation gives a practical boundary: lastmod should describe the last significant modification, while a minor change such as a copyright date should not trigger it.

Editorial teams need this distinction because a cosmetic date reset creates no new proof. It may also corrupt the internal record: if every small copy edit becomes an “update”, a researcher cannot tell when the underlying statistic, expert role or service description was last checked.

Material revisions commonly include:

• Replacing an older statistic with a new result from the primary publisher and naming the new reference period.
• Correcting an expert's title, organisation or remit against a verified profile.
• Updating markets served, regulatory requirements, product availability or delivery steps.
• Replacing a broken or secondary citation with a live primary document.
• Adding a substantial answer section and aligning internal links, Article markup, dates and image descriptions.

Spelling corrections, punctuation edits and sentence reordering are maintenance. They should be logged if the workflow requires it, but they do not normally justify a new public date. A joined SEO and PR evidence workflow makes this policy enforceable across teams.

How does a 90-day evidence refresh cycle work?

A 90-day evidence refresh cycle inventories claims in days 1–30, revises priority pages in days 31–60 and verifies recrawling plus answer performance in days 61–90. The cycle is an operating rhythm, not a universal expiry date for every fact.

The first month covers more than articles. Service pages, case studies, executive biographies, editorial coverage, research pages and FAQ answers all contribute to the public evidence environment. Each important URL should have a named owner, a main claim, an original source, a source date and a last-verification date.

The second month concentrates effort according to risk. Claims that affect medical, legal, financial or purchasing decisions deserve shorter review intervals than stable corporate details. The third month aligns the technical signals, observes crawler access and repeats the same answer-engine tests used before the revision.

A workable refresh register contains:

• Claim owner: the expert or team authorised to confirm the fact.
• Volatility band: monthly, quarterly, twice-yearly or annual review.
• Revision record: the previous statement, new statement and primary proof for the change.
• Technical record: dateModified, sitemap lastmod, canonical URL, internal links and crawl status.
• Answer record: brand mention, factual accuracy, cited URL and query context.

This register prevents content production from becoming a volume contest. Instead of opening a near-duplicate URL every week, the organisation improves the page that already carries subject authority. The principle matches the need to build PR and SEO around shared outcomes, not separate publishing quotas.

How should datePublished, dateModified and sitemap lastmod align?

datePublished should preserve the first publication date, dateModified should identify the latest material editorial revision and sitemap lastmod should signal that same significant change for crawl scheduling. They serve different systems but should originate from one verified change record.

Google recommends ISO 8601 dates with a timezone for Article structured data. Its sitemap guidance says an accurate and consistently verifiable lastmod value can help schedule recrawling, while changefreq and priority are ignored. The implementation should therefore favour honest change data over a collection of decorative fields.

Consider a page that receives a punctuation fix on 12 August. Its public modified date and sitemap record should remain unchanged. If the team replaces the core study with 2026 findings, rewrites the interpretation and removes obsolete guidance on the same day, both the public modified date and lastmod can legitimately show 12 August.

A shared publication form reduces errors. It should capture why the page changed, which sections changed, the sources used, the approver and the exact deployment time. Editors, developers and PR leads can then work from the same record instead of sending search systems conflicting dates.

How should refreshed pages be retested in LLM answers?

Refreshed pages should be retested with a fixed set of at least 20 intent-matched questions, with every question run three times. One favourable ChatGPT or Perplexity response cannot establish that the revision changed source selection.

Start with access. Confirm a 200 response, a correct canonical URL, relevant crawler permission and an updated sitemap value. OpenAI advises publishers that OAI-SearchBot access is necessary for inclusion in ChatGPT search results, while also making clear that no position is guaranteed. Server logs can then show whether the new version was requested.

Next, test three query families:

• Definition questions that ask what a method, service or category is.
• Decision questions that ask which criteria should guide a supplier or process choice.
• Verification questions that ask for the source, date or proof behind a brand claim.

For every answer, record whether the organisation is mentioned, whether the subject association is correct, whether key facts are accurate and which URL is cited. Keep engine, market setting and test window consistent between the before and after samples. A crawl observation proves access; a citation proves that the answering system selected the source for a particular request. Combining them into one KPI hides where the process failed.

This distinction reflects brand authority in the AI era: discoverability matters, but authority is expressed through relevant, corroborated use of the source.

How should a GEO agency prioritise refresh work?

A GEO agency should prioritise refresh work using four factors: business risk, claim volatility, existing citation exposure and observed recrawl interval. The highest-traffic URL is not automatically first; the page capable of causing the greatest decision or reputation error when wrong deserves faster action.

Each factor can be scored from 1 to 5. A time-sensitive medical claim might receive a 5 for risk, while a stable office address receives a 1 for volatility. A page that appears frequently in AI answers but now relies on old evidence also receives a high exposure score because an inaccurate claim has a wider path to repetition.

The resulting work is multidisciplinary. PR specialists source a current expert view or independent evidence. Editors turn it into a concise, answerable structure. Technical teams publish the crawl and date signals. Measurement teams repeat the query set and compare source behaviour. No single “refresh” button can complete all four jobs.

A monthly report should therefore count verified claims, corrected contradictions, recrawled priority URLs and pages that retained or recovered relevant citations. Those measures show whether the organisation improved its evidence system. A raw total of newly published posts does not.

Frequently Asked Questions

The core decisions are to match review frequency to claim volatility, reserve modified dates for material revisions and measure crawl access separately from citation outcomes.