How Often Should You Update Content for AI Search? The 2026 Refresh Cadence Playbook

Last updated: 15 August 2026 · LeadsNow AI, Melbourne

Every AI search guide tells you freshness matters. Almost none of them tell you what to actually do on a Tuesday morning: which pages to open, what to change, how often to come back, and how to prove to the engines that anything happened. That gap is what this guide fills — a working refresh cadence you can run without hiring anyone.

It matters more than it used to because the buyer journey has moved. B2B buyers increasingly start their research inside AI chatbots, and those engines make a judgement classic Google rarely made this bluntly: is this page current enough to trust with a recommendation?

How stale is too stale? What the data says

The best public evidence comes from AirOps’ 2026 State of AI Search report, which analysed citation patterns across ChatGPT, Perplexity, Gemini and Google AI Mode. The pattern is blunt:

• More than 70% of all pages cited by AI had been updated within the past 12 months, and more than 50% within six months.
• On commercial queries it tightens further: about 83% of citations go to pages updated within the past year, and more than 60% within six months.
• Pages not updated quarterly are over 3x more likely to lose visibility than recently refreshed pages — and in fast-moving verticals like SaaS and finance, AirOps reports pages older than three months see steep drops in citation likelihood.

One caveat, in the interest of honest sourcing: the public report page does not disclose sample size or query methodology, so treat the percentages as directional rather than gospel. Directionally, though, they match what we see in our own citation monitoring — the pages engines fetch and cite from our site are overwhelmingly the ones we have touched recently.

Read that as a decay curve, not a cliff. Your page does not die on day 91; it slides down the shortlist a little more every quarter you ignore it. And because freshness is only one of the filters engines apply — we unpacked the rest in how to be in the 15% of pages ChatGPT actually cites — a stale page loses twice: it is less likely to make the shortlist, and easier to beat once it is there.

Refresh cadence by page type

Not every page earns the same attention. Here is the rotation we run on our own site and recommend to clients:

The intervals are deliberately aligned with the data: quarterly keeps your money pages inside the six-month window where over 60% of commercial citations live — and clear of the over-3x visibility-loss penalty AirOps measured for pages not updated quarterly. Six-monthly keeps everything else safely inside the 12-month window. Two rules for using the table: the interval is a floor, not a ceiling, and a trigger event always beats the schedule. If the regulator, the source study or the vendor’s pricing page changes on a Tuesday, the refresh happens that week — not at the next calendar review.

What actually counts as an update

Engines are pattern-matching on substance, not on a changed timestamp. A refresh should leave the page measurably different. In priority order:

1. Statistics and their sources

Replace superseded numbers, and re-verify the ones you keep — a 2024 stat with a live source beats a 2026 stat with a dead link. If a claim can no longer be verified, delete it rather than let it rot on the page.

2. Dates and year references

Sweep the copy for “in 2025”, “as of last year”, “the current version”, and year-stamped titles. Nothing signals staleness to a model faster than a page that says “2026 guide” over paragraphs describing 2024.

3. Your answer capsule

The summary block at the top of the page is the passage most likely to be quoted, so it ages fastest. Rewrite it against the question as buyers ask it now, with this quarter’s numbers in it.

4. FAQ answers

Re-read every FAQ answer as if a chatbot were about to quote it verbatim today. Update the visible answer and the matching FAQPage schema — if they drift apart, you are feeding engines two versions of the truth.

5. Examples, screenshots and internal links

Swap dated examples for current ones, re-shoot screenshots of interfaces that have changed, and link newer sibling pages published since the last pass. This is also where refresh work compounds with authority work — a freshly updated page is a better landing spot for the earned media and digital PR that gets you retrieved in the first place.

Proving the update happened: dateModified and visible dates

Once the substance is updated, make it machine-readable. Google’s guidance on publication dates is the cleanest checklist going:

• Use Article (or another CreativeWork) structured data with datePublished and dateModified fields.
• Show a user-visible date, prominently, with a label — Google’s own wording is to label dates “with text like ‘Publish’ or ‘Last updated'”. A date only in schema is a whisper; put it at the top of the page.
• Keep visible and structured dates identical. Google explicitly asks that the user-visible date and the structured-data value match — and that dates describe the actual update of the page, not future dates or the events discussed on it.
• Minimise other dates on the page so parsers do not pick up the wrong one.

The order matters: change the content, then the dates. Bumping dateModified on an untouched page fails Google’s own requirement that the date describe a real update — and engines that can diff crawls can see nothing changed. Google News policy goes further still: its Publisher Center date guidelines tell publishers not to “artificially freshen” a story without adding significant information, and that republishing an old piece under a new URL with a new date is against the rules. A fresh date is only warranted when the page has substantially changed — and the Wayback Machine keeps receipts either way. The test we apply internally: could a reader diff the old and new versions and point at what improved? If not, the date stays where it was.

Prioritise like a pipeline, not a librarian

If you have 200 posts and a spare afternoon a fortnight, do not refresh chronologically. The best triage signal is already sitting in your server logs: which pages the AI crawlers and fetchers — GPTBot and ChatGPT-User (OpenAI), PerplexityBot, ClaudeBot, Google-Extended and friends — are actually requesting. Those logs tell you which URLs the engines already consider fetch-worthy, and the fetches cluster hard on specific pages; that concentration is your priority list. Triage:

1. Pages bots fetch that carry commercial intent. These are live citation candidates on the queries that make money, and the freshness penalty is steepest exactly there — AirOps’ tightest windows (about 83% within a year, over 60% within six months) were on commercial queries. Service pages, pricing-adjacent pages and comparison posts jump the queue, on the quarterly cadence, regardless of how recently you published them.
2. Pages bots fetch that have gone stale. An AI engine requesting a page you have not touched in 18 months is an open goal: it is actively considering a document you have let rot. These are the fastest wins in the whole programme.
3. Then pages already earning citations. Protect what is working before rescuing what is not. If you are tracking share of answer on your revenue-critical prompts, your defend-list writes itself: any cited page approaching six months since its last touch gets refreshed this cycle.
4. Then high-potential stale pages. Pages that used to earn traffic or citations and have gone 12+ months untouched but no longer attract bot fetches are usually a decay story — still the highest-leverage rescues, because the substance often needs an hour, not a rewrite. Refresh substantially or consolidate into a stronger page.
5. Last, everything else — and be honest about pages not worth maintaining. Consolidate or retire them; a smaller library you can keep fresh beats a big one going stale in parallel.

No log access? Second-best proxies: pages with existing citations in your AI-visibility monitoring, then pages with impressions on commercial queries in Search Console.

If you would rather have that triage done for you against your actual pipeline — which pages feed appointments, which are ageing out — Book a call and we will walk through it on your site.

The refresh checklist

Run this per page, roughly 45–60 minutes for a typical post:

1. Re-read the target question. Is the page still answering what buyers ask in 2026, in their words?
2. Rewrite the answer capsule with current numbers and this quarter’s framing.
3. Verify every statistic at its primary source; replace or delete anything that fails.
4. Sweep for year references, “recently”, “last year”, and outdated product or pricing details.
5. Update FAQ answers — visible text and FAQPage schema together.
6. Refresh examples and screenshots; fix dead links; add internal links to pages published since the last pass.
7. Add one genuinely new element — a new question answered, a new data point, a new section — so the update adds value rather than just polishing.
8. Set dateModified and the visible “Last updated” line to today, matching exactly.
9. Log the pass (page, date, what changed) and diarise the next one per the cadence table above.

That log is the difference between a cadence and a good intention. A spreadsheet with four columns — URL, page type, last refreshed, next due — is enough to run the whole program.

Where this fits in a lead-generation engine

We treat refresh cadence as pipeline maintenance, not housekeeping, because our own demand depends on being the source AI engines quote when Australian businesses ask about lead generation and appointment setting. The same discipline — quarterly money-page passes, six-monthly guide passes, dates that always tell the truth — sits underneath 50,769+ AI-booked sales appointments since 2017 and 1M+ leads generated, with 25 filmed client case studies behind it.

Want your refresh priorities mapped for you? We will audit which of your pages are going stale on revenue-critical questions, what each one needs, and what fixing them is worth in booked appointments. Book a call.

Frequently asked questions

Does changing the date on a page without editing the content help with AI search?

No — and it can undermine trust in your dates entirely. Google’s publication-date guidance requires that dates “describe the publication or update date of the page” and that user-visible dates match the structured-data values, and Google News policy explicitly tells publishers not to “artificially freshen” a story without adding significant information. A bumped dateModified over an unchanged page is a date that describes nothing, and crawl-to-crawl comparison makes that visible. Change the substance first; the date is just the receipt.

How much do I need to change for a refresh to count?

Enough that a before-and-after diff shows real substance: updated statistics, a rewritten answer capsule, corrected details, at least one new element the page did not have before. A useful internal bar is that you could honestly write a one-line changelog entry for the pass. Fixing a typo and re-saving does not clear that bar; updating three stats, the capsule and two FAQ answers does.

Should I show the publish date or the last-updated date?

Show a clearly labelled “Last updated” date prominently near the top, and keep datePublished and dateModified both in your Article schema. Google’s guidance is to feature a visible, labelled date and keep it consistent with the structured data — and to minimise other stray dates on the page so parsers select the right one.

Is it better to refresh old posts or write new ones?

For pages targeting queries you already care about, refresh first — an existing page with history that you bring back inside the six-month window is usually faster to re-cite than a brand-new URL is to establish. Write new posts for questions you have never answered; refresh the ones you have. Most content operations get the ratio backwards and keep publishing while their money pages quietly age out of the citation pool.

Which pages should I refresh first in a large back-catalogue?

The ones AI bots already fetch. Filter your server logs for AI user agents (GPTBot, ChatGPT-User, PerplexityBot, ClaudeBot and similar), then prioritise fetched pages with commercial intent, followed by fetched pages that have gone stale. Without log access, use your AI-visibility monitoring and Search Console commercial-query impressions as proxies.

How do I know if my refresh program is working?

Measure share of answer, not effort: pick the prompts that drive revenue, record which sources the major engines cite before you start, then re-check monthly after each refresh cycle. Citations moving on refreshed pages — and holding on defended ones — is the signal. We score it one step further downstream, in booked appointments from AI-referred visitors, because that is the number the cadence exists to protect.