How fresh does a page have to be for AI search?

How fresh does a page have to be for AI search?

Fresh enough to be accurate, and no fresher. Google's documentation does not describe a recency threshold that pages must clear. It describes systems that retrieve relevant, up to date pages from the Search index. Accuracy is the requirement. A recent date is not.

I get asked this constantly, usually by someone who has been told to republish their whole blog every quarter. That advice costs real money and rests on an instinct rather than on anything a search engine has published.

So let me separate what is documented from what is folklore, because the gap between those two is where most content budgets go to die.

What does Google actually say about freshness?

Google's guide to optimizing for generative AI features describes retrieval augmented generation, which it also calls grounding, as a technique used to improve the quality, accuracy and freshness of AI responses by relying on core Search ranking systems to retrieve relevant, up to date web pages from the Search index.

Read that carefully. Freshness is a property Google is trying to achieve in its answers by choosing which pages to retrieve. It is not a bar your page is being asked to clear. The system is selecting for currency of information, which is a different thing from recency of publication.

Google is also explicit that this is not a separate game. Its documentation says the best practices for SEO continue to be relevant because the generative AI features on Search are rooted in the core Search ranking and quality systems. Its AI features documentation goes further and states there are no additional requirements to appear in AI Overviews or AI Mode.

Google even defines the vocabulary in that same guide, noting that AEO stands for answer engine optimization and GEO for generative engine optimization. Useful, because plenty of people selling AEO services imply a mechanism Google has never described.

Why does recrawl cadence matter more than your publish date?

Because a change nobody has crawled does not exist yet. Google's documentation says crawling can take anywhere from several days to several months, depending on how often its systems determine a page needs to be refreshed. Your update is only real once that happens.

This reframes the whole problem. The question is not how recently you changed a page. It is how recently Google looked. Those two things are connected but they are not the same, and only one of them is under your direct control.

It also explains something that confuses people. You update a page on Tuesday, check an AI answer on Wednesday, and nothing has changed. That is not evidence that the update failed. It is evidence that the update has not been seen. I have written more about how crawlers decide how often to come back, and the short version is that they reward sites that actually change.

Does updating the date on a page help?

Changing a displayed date without changing the content is not a strategy, it is a lie with a timestamp. Nothing in Google's documentation suggests a date alone earns anything, and a reader who notices will trust the page less.

The same applies to bulk republishing. Rewriting the publish date across a few hundred posts in one afternoon tells a crawler that your whole archive changed simultaneously, which is not true and not useful. It also destroys your ability to tell which updates actually mattered.

Dates in machine readable places get the same treatment. There is a persistent belief that a sitemap value can be used as a lever, which I have picked apart in what a sitemap lastmod date actually does. Signals describe reality. They do not create it.

Which pages genuinely need to be fresh?

Pages where the correct answer has changed. Pricing, platform behaviour, legal or compliance guidance, tool comparisons, anything naming a version, and anything where a reader acting on last year's advice would get a worse outcome.

Everything else can sit. An explanation of how internal linking works does not expire. A framework for choosing a channel does not expire. If the underlying truth has not moved, republishing the page adds nothing except a false signal of activity.

The practical filter I use is a single question: if a reader followed this page today, would anything go wrong? If the answer is no, leave it alone. If the answer is yes, it is urgent, and it should have been fixed before anyone asked about freshness.

There is a sharper version of this problem when the stale page is the one an assistant keeps quoting about your own company. Finding and correcting that specific source is a different job from a refresh programme, and I set it out in why AI answers about your company are out of date.

What does query fan-out mean for an ageing page?

It means a single page can be pulled into answers you never targeted. Google describes query fan-out as a set of concurrent related queries generated by the model to fetch additional relevant results, giving the example of a lawn full of weeds expanding into queries about herbicides, removing weeds without chemicals and preventing weeds.

The implication is that coverage and correctness across a topic matter more than the recency of any one page. If your article is the best answer to one of the fanned out queries, it can be retrieved even if it is not the thing the person originally typed.

That is an argument for depth, not for churn. An old page that answers a sub question precisely is more useful to this mechanism than a recently republished page that answers the headline question vaguely. It is also why the way a page is structured into discrete, self contained answers matters so much, which I have gone into in why AI search tends to quote one paragraph.

What does a real refresh look like?

A real refresh changes claims, not cosmetics. You reverify every checkable fact, delete what is no longer true, add what has actually changed, and only then touch the date. If nothing in the first three steps produced an edit, there was nothing to refresh.

I run my own archive that way. With more than 350 articles published, I cannot and should not update all of them. I update the ones where a platform changed, a price changed, or I now think I was wrong. The rest stay exactly as they were, because they are still correct.

The uncomfortable part is deletion. Some pages should not be refreshed at all, they should be removed or consolidated. That is a harder decision than rewriting an intro, which is exactly why most people rewrite the intro instead.

Do the eligibility basics still matter more than any of this?

Yes, and by a wide margin. Google's AI features documentation states that to be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements.

That single sentence outranks every freshness tactic. If a page is not indexed, nothing else you do to it matters. If you have suppressed previews, you have opted out of exactly the thing you were trying to win. Google names the relevant preview controls directly: nosnippet, data-nosnippet and max-snippet.

I check those before I look at anything else on a client site, because I have found suppressed previews on pages whose owners were paying for visibility work. It is the cheapest possible fix and it is invisible until you go looking. Confirm current behaviour in Google's own documentation, since these controls and their handling do get updated.

How do you tell whether a refresh worked?

Give it long enough to have been crawled, then compare like with like. Because recrawl can take days or months, judging a refresh after a week tells you nothing except that you are impatient.

What I look for is whether the page now gets pulled into a wider set of related questions, not whether one specific answer changed overnight. Fan out means the useful signal is breadth of retrieval, and breadth moves slowly.

I also keep a note of what I changed and why. Without that, a refresh programme becomes unfalsifiable, and unfalsifiable programmes run forever because nobody can prove they are not working.

What should you do next?

Stop scheduling refreshes by date and start triggering them by change. Make a short list of the pages whose correctness depends on something outside your control, and check those on a real cadence. Leave the rest alone until a fact moves.

Then go and confirm your important pages are indexed and allowed to show a snippet. That is a one hour job and it matters more than any freshness plan you could write this quarter, because it is the documented eligibility requirement rather than an inference about one.

If you want a second opinion on which pages in your archive are genuinely stale and which ones are just old, reach out. Telling those two apart is most of the work.