Content freshness affects SEO only for queries that demand current information — an old, well-maintained page is not treated as low quality by default, but an unmaintained or inaccurate site is. Google’s own rater guidelines state that “stale” pages can have high Page Quality ratings, while “unmaintained/abandoned” content is a reason for a low one. So the decision to refresh or write new comes down to two things: whether the queries your page targets actually demand recency, and whether refreshing would make the content genuinely current rather than just newly dated. Everything below is separated into what Google’s documentation says, what practitioners report, and what falls into neither.

What does Google actually say about freshness?

The most specific public statement Google has made about freshness is section 18.0 of the Search Quality Evaluator Guidelines (2025-09-11 version, accessed 2026-10-29) — and the honest framing of that source matters before the quotes. The Guidelines instruct human quality raters who score sample results; Google uses that rater feedback in training its systems. So the document is not a ranking algorithm specification, but it is Google’s most detailed public description of how query quality is judged, and it is the closest thing to primary documentation on freshness that exists.

That section says that “some queries demand very recent or ‘fresh’ information,” and lists four query types where “only pages with current, recent, or updated content should get high Needs Met ratings”:

Query typeExamples from the GuidelinesWhat “fresh” means for your content
Breaking newsa severe-weather event currently happeningYour page must report the current state; a well-written but last-week version cannot get a high rating for the live event
Recurring events (elections, sports, TV shows, conferences)an upcoming election, this season’s tournament, the current tax-year’s formsFresh means the current cycle: last year’s forms page does not answer this year’s query, however well it was built
Current informationa city’s population, today’s airfare between two airportsFresh means the number or price as of now; the page is useful only while its figures track reality
Product queriesa phone brand nameGoogle says to “assume users are looking for information about the most recent model/version” — a review of the previous generation answers a different query than the bare product name

The second half of the section is the part most “freshness” advice misses. Freshness is “generally less of a concern for Page Quality rating,” and the Guidelines say plainly: “Stale” pages can have high Page Quality ratings. A page that is old but still accurate, complete and well-maintained is not demoted for its age in the quality judgment.

The caveat sits in the same section, and it is what real-world “my old site stopped ranking” stories usually point at: “unmaintained/abandoned ‘old’ websites or unmaintained and inaccurate/misleading content is a reason for a low Page Quality rating.” The distinction is between old and still correct and old and visibly falling behind. The first is fine; the second is a quality problem that looks like a freshness problem from the outside.

When does freshness actually matter for a query?

Practically: freshness matters when the answer to the query has a date on it. The four types above share one property — the query implicitly means now, and a page that describes last year’s state of the world answers a different question. If your page targets any of those query shapes, its content needs to be current or it will underperform, regardless of how well it was written.

If your page targets evergreen questions — how something works, how to do something that hasn’t changed, what a term means, comparisons whose substance is stable — the Guidelines’ own language is that freshness is “generally less of a concern.” There is a nuance worth keeping: a query can look timeless and quietly be current. “What’s the population of X” is a current-information query, not an evergreen one; “how do elections work” is evergreen, but “this year’s election” is a recurring-event query. The test is what the user expects the answer to be today. If your answer would be unchanged in five years, age is not a problem. If it would be wrong, it’s a freshness problem in disguise.

And the evergreen side still carries the unmaintained-site risk. A blog whose numbers, product versions, examples and links all point to 2021 is the “unmaintained and inaccurate” case from section 18.0 — not because of dates per se, but because the accumulated inaccuracy is a quality signal. Freshness for evergreen content is therefore less a ranking input than a maintenance discipline: the page stays high-quality by staying correct.

Should you refresh an old post or write a new one?

There is no primary study comparing refresh-versus-new performance, and the claim you will see in the practitioner community — “updated beats new for informational content” — is experience reports, not measurements. We label it that way here too: it is practitioner consensus, observational, and no published dataset establishes cause and effect. That said, the signal is real enough to take seriously in the one place where it is documented with numbers: in the September 2026 spam update study of roughly 100 SMB sites, pages updated in the prior 30 days saw clicks rise 23% while the portfolio median fell — an observational pattern in one self-selected portfolio, not a guarantee, but consistent with the same rule.

The standard practitioner method for the decision is to let Search Console pick the candidates instead of guessing:

  1. Find pages with declining impressions in the GSC Pages report. A page whose impressions slide while positions hold is often a page whose content stopped answering the query well — the first candidate to check.
  2. Find page-2 pages that could plausibly win. A page consistently ranking around positions 4–10 for a query it genuinely answers well is the classic refresh candidate: the relevance is already recognized, and the gap is often depth or currency.
  3. Decide by intent, not by page. If the current state of the query is new enough to be a different question — a new product generation, a new rules cycle, a workflow that changed — a new post is the honest answer, and the old one gets a link or a redirect, not a facelift. If the question is the same but the substance inside it went stale, refresh the existing page.

What a real refresh means, by practitioner consensus and by the Guidelines’ own logic: make the content actually current. Update the facts, numbers, dates, examples and versions; fix what is now wrong; add what is new. Not bump the published date. If the page was AI-drafted, that is where the distinction matters most: re-prompting a model on stale input is not a refresh, and the review discipline from how to use AI-drafted content safely applies to the refreshed page exactly as it does to a new one. The SQEG anticipates exactly this shortcut — “the date the page was created may be different from when the content was last updated or modified,” and “some websites always show the current date, regardless of when the content was last updated” — which is a rater-side warning that displayed dates are treated as claims about the content, not the other way around.

One structural point makes refreshing cheaper than it looks: if the page sits in a content cluster, the refresh stays inside the cluster. The internal links that made the page discoverable for its sub-question keep working, the pillar still points to it, and you update one page rather than rebuilding a topic. The alternative — writing a “2027 edition” as a new post while the old one stays live — is how clusters develop keyword cannibalization: two pages for one query, one of them stale.

Does changing the date or the lastmod tag do anything?

No more than any other metadata, and the accurate statement matters here. Google’s sitemaps documentation (last updated 2025-12-10) says sitemaps provide “valuable information about these files,” such as “when the page was last updated” — lastmod is a hint that helps Google prioritize crawling, not a freshness pass or a ranking signal. Setting lastmod to today on a page you did not touch is not a tactic; at worst it is wasted, and at best the crawl happens and the content is judged on what is actually in it.

The same logic applies to the dates readers see. Since the Guidelines treat displayed dates as claims, the honest pattern is the one that costs nothing: change the date only when the content actually changed, and say what changed in a short “updated” note. A visible “Reviewed 2026-10” line does real work for readers and for the AI-answer side of the coin, too — current, verifiable content is one of the mechanics that makes a passage AI-citable, and a model composing an answer about a moving target has no way to verify a page whose last visible update was three years ago. That is our inference from how answers are grounded in citable passages, not a documented Google mechanism; the documented part is what section 18.0 says above.

Where to start

Pick the five query families that matter most to revenue, classify each against the four types — current-information, recurring-event, breaking-news, product, or evergreen — and treat the classification as the decision engine: pages whose queries demand recency get their content made genuinely current now; evergreen pages get a correctness pass (wrong numbers, dead examples, old versions fixed) and a moving date when — and only when — that pass happened. If you want this run as an audit rather than a judgment call — the query classification, the GSC impression trends, the refresh list with the change log so you can tell later whether it worked — that is exactly what we run end to end, and the full process is in our methodology.