If your blog has around 30 posts and you publish weekly, the honest split between refreshing and publishing is not a ratio you decide in advance — it is the output of one rule. Refresh a post when it has lost roughly half its impressions over a rolling 90-day comparison and the search results still want the kind of page you wrote. Publish new only into clusters you don't already cover. On a library that size, that rule usually surfaces two or three refresh candidates per quarter, which means most of your slots still go to new posts — the opposite of what the popular advice implies.
That advice deserves a closer look, because the numbers behind it were measured somewhere very different from your blog.
The foundational number in every "update your old posts" argument comes from HubSpot's historical optimization programme: 76% of monthly blog views and 92% of monthly blog leads came from posts published before that month, and 46% of monthly leads came from just 30 individual posts. Updating and republishing lifted monthly organic search views on those posts by an average of 106% and tripled their leads.
Impressive. Now read the scope sentence in the same article: "we publish about 200 new posts every month, and at that time, we had accumulated nearly 6,000 total posts on our blog."
Do the arithmetic. At 6,000 posts and 200 a month, roughly 3% of HubSpot's library was new in any given month. Of course 76% of views came from old posts — 97% of the posts were old. The statistic is close to a tautology once a blog is large enough, and it says almost nothing about where the marginal hour should go on a blog of 30 posts. On your blog, four new posts a month is about 13% of the library. Same arithmetic, wildly different answer.
None of this makes refreshing wrong. It makes the reason wrong, and the reason determines the rule. Refresh because "76% of traffic comes from old posts" and you'll rewrite posts that never had traffic to lose. Refresh because a specific page used to earn impressions and stopped, and you're fixing something actually broken. Which raises the practical question: on a 30-post blog, what can break?
Blog traffic follows a power law, and it does so brutally. Green Flag Digital's October 2025 analysis of real B2B blogs found the top 10% of posts carried 81.4% of blog traffic at Wiz (429 posts) and 98.2% at Abnormal.ai (387 posts). Even NerdWallet, with 8,531 posts, concentrated 76.9% of traffic in its top 10%.
Apply that shape to 30 posts and your top 10% is three posts. Perhaps five or six more get meaningful impressions. The rest never ranked, which means they cannot decay — they never started. Internalise that before you open Search Console: a refresh rule keyed to lost impressions returns a very short list on a small blog, and that short list is the answer, not a broken rule.
Decay itself is real. Animalz coined the framing in 2018: "Content decay is the natural loss of relevancy experienced by nearly all well-performing posts." The best measured baseline comes from Workshop Digital's 2.5-year analysis, unusual in having an untouched control group: control pages declined 20.2% in organic sessions over matched 90-day windows, with blog-only control pages down 9.1%. That is the honest size of "do nothing" — noticeable, not catastrophic, and nowhere near the round scare numbers circulating in SEO roundups.
But not every traffic decline is decay, and this is where most small-blog refresh work goes to waste. Search Engine Land's July 2026 breakdown maps Search Console signals to four distinct causes, and only one of them is fixed by rewriting:
| What you see in GSC | Diagnosis | What to do |
|---|---|---|
| Clicks down, impressions down, position worse | Ranking decay | Refresh substantively |
| Clicks down, impressions flat or up, position stable | Zero-click capture | Not a content problem |
| Clicks down, position roughly holding | Intent drift | Reformat to match the current SERP |
| Clicks down, impressions down, position held or improved | Demand decay | Usually nothing |
The article's line on the second row is the one worth taping to your monitor: "You didn't lose rankings or quality. An AI Overview is now answering the query on the results page." And on the fourth: "No rewrite resurrects a search people stopped running."
That table is why the rule below is written in impressions rather than clicks.
Refresh a post when its impressions have fallen by roughly half over a rolling 90-day window compared with the previous 90 days, and the current search results still reward the format you published.
Two clauses, and both are load-bearing.
The impressions clause exists to separate real ranking loss from AI Overview capture. Clicks alone mislead: a page can hold position one, keep every impression it ever had, and still lose half its clicks because an AI summary now answers the query above it. Seer Interactive's study of 53 brands across 5.47 million queries found that a page cited inside an AI Overview gets about 2.1% CTR, versus 0.9% when it isn't cited and 3.3% with no AI Overview at all. Rewriting won't reverse that. Impressions holding while clicks fall is the tell.
The intent clause exists because a refresh only works if the SERP still wants your kind of page. Ahrefs tracked the results for "how to start a blog" over a year and logged 13 SERP changes with a 31/100 similarity score — Reddit threads went from 34% to 60% of page one while how-to guides fell from 27% to 22%. If your competition is now a forum thread, expanding your guide by 40% is not the move.
Why half, when Ahrefs suggests triaging anything down 20% year over year and Animalz uses 20% over 90 days or a five-position drop? Because those thresholds were calibrated on libraries with enough volume to make 20% a signal. On a page doing 400 impressions a quarter, a 20% swing is a seasonal wobble, a core update settling, or one competitor's launch week. Fifty percent is still noisy on a small base — but it's noise you can afford to act on two or three times a year.
Before you run this comparison, one warning that almost nobody writing on this topic is giving. Google stopped supporting the &num=100 search parameter in mid-September 2025. Until then, rank-tracking crawlers pulling 100 results per query had been inflating Search Console impression counts — producing what Karen Bellin described as an "alligator effect" of rising impressions against steady clicks from roughly February through September 2025.
The consequence for you, in August 2026: any impressions comparison that straddles September 2025 is contaminated. Search Console's 16-month lookback — named as a limitation in Workshop Digital's own study — still reaches back into the inflated window. Compare "this quarter versus the same quarter last year" today and you may see a large impressions drop on a page that never actually declined.
Use rolling 90-day versus previous 90-day comparisons entirely inside post-September-2025 data. When a case looks ambiguous, open the live SERP and look.
Say a post passes both clauses. What you do next matters more than most people assume, and there is finally controlled evidence on it.
RepublishAI's March 2026 study compared 6,819 updated URLs against 8,168 never-updated controls — 14,987 URLs across 20 verticals over a roughly 76-day window — and grouped results by how much of the page actually changed. Average position change: 0–10% content change, −0.51; 11–30%, −2.18; 31–100%, +5.45; never-updated control, −2.51. Only the 31–100% bucket reached statistical significance (p=0.026, a net +7.96 positions versus control).
Read those buckets carefully, because secondary write-ups overstate them. Small and medium edits didn't hurt; they landed in the same range as doing nothing at all. The defensible claim is that anything short of a substantial rewrite was statistically indistinguishable from leaving the page alone. Which means a half-hearted refresh costs you a publishing slot and buys you a rounding error.
The study's own limitations deserve repeating: correlation isn't causation, the sample was biased toward URLs already ranking in the top 100, and content-change detection via scraping can miss edits.
Google's guidance points the same way. Its core updates documentation asks publishers to improve content in "meaningful ways" — "rewriting or restructuring your content". On dates, Google's documentation says a page's date should reflect "the publication or update date of the page", and that Google "examines several factors" to estimate when a page was significantly updated rather than trusting your declaration. John Mueller has said plainly that changing publishing dates won't improve rankings without significant content changes. Or, as the Search Engine Land breakdown puts it: "Adding 300 words won't move it."
So what does a real refresh contain? The recurring shape across every credible source:
Two rules never break: keep the original URL, and add a visible editor's note about what changed. Zapier has updated its time-tracking software roundup every year since 2018 on the same URL and it still ranks first for its target keyword. Orbit Media has kept the URL of its annual research post for a decade.
Step four quietly eats an afternoon. A two-year-old post's citations are the most likely thing to have rotted — sources move, studies get republished at new paths, vendors retire pages. That's maintenance work rather than writing work, which is why it gets skipped, and why link integrity is far cheaper to enforce when a post is drafted than to repair a year later.
The case for the other half of the allocation is stronger than refresh advocates admit — with a timing wrinkle that makes it manageable.
New pages are a long shot. Ahrefs' May 2025 restudy of a million randomly sampled URLs found that only 1.74% of newly published pages reach Google's top 10 within a year, down from 5.7% in the 2017 original; under a stricter filter limited to non-empty English content, 6.11%. Meanwhile 72.9% of top-10 pages are more than three years old, and pages under a year old make up just 13.7% of top-10 results.
That sounds like an argument for never publishing again. It isn't, because of one more figure in the same dataset: 40.82% of the pages that did reach the top 10 got there within a month, and prospects diminish significantly after six months.
That asymmetry is the gift. A new post tells you fairly quickly whether it has a pulse. Publish, wait 30–60 days, and check whether it picked up any impressions at all. If it did, it's a future refresh candidate — a lottery ticket you can top up later. If it's flat at 60 days, the honest read is usually that you missed the intent or picked a keyword you can't reach yet. Either way the lesson is cheap, and the split stops being a fixed ratio and becomes a feedback loop.
The condition on publishing is the harder half: publish into clusters you don't already cover. A second post on ground you already occupy is not coverage, it's competition with yourself — it splits internal link signals and forces search engines and AI systems to pick a representative page from near-duplicates. We've written a full breakdown of what keyword cannibalization actually costs a small blog, and the summary is that prevention is far cheaper than a consolidation audit later.
So define "uncovered" concretely. A cluster is covered when one page answers its core question directly, earns at least some impressions for a query in that theme, and is the page you'd link to when the topic comes up. If you'd struggle to name which post owns a question, that's a gap. If two posts both plausibly own it, that's a consolidation job. However you make that call, you need something that remembers what you've already written and planned — Magic Share keeps a topic ledger and shortlists 6–10 scored ideas per run so a new draft lands in a gap rather than on top of an existing post.
Everything above would have been roughly true in 2023. What changed is where citations come from.
Seer Interactive's July 2026 study analysed 7,683 dated pages carrying 47,097 citations across ChatGPT, Gemini and Perplexity, and found that 75% of cited pages had been updated within the past year and 88% within two years. Content older than three years was essentially absent. But the headline finding is the one that reframes this entire post: among pages that appeared fresh, 72% had been recently updated while only 42% were originally published within the past year — more than 26% were first published over two years ago.
Seer's summary: "The freshness LLMs reward is being manufactured by updates, not by new publishing." And: "Publish and forget loses. Publish and maintain wins."
Keep the caveat attached, though. Ahrefs' analysis of nearly 17 million cited URLs found AI-cited content averages 1,064 days old versus 1,432 days for organic top-10 results — 25.7% fresher, with the author's note that "like traditional search, AI assistants still prefer citing long-lived content." Nearly three years old, on average. Freshness is a tiebreaker among established pages, not a bias toward new ones — which is why an updated three-year-old post is the strongest asset here, and a brand-new one the weakest.
If AI citation is where you're aiming, the page-level craft matters as much as the update date; our guide to earning citations from ChatGPT and AI Overviews covers the paragraph-level patterns that get extracted.
No one should hand you a 60/40 split, and this post won't. The split is an output.
Run the audit quarterly — on a 30-post blog it takes about half an hour in Search Console. Compare rolling 90 days against the previous 90, page by page, reading clicks, impressions and average position together. Most quarters you will find two or three posts that pass both clauses of the rule. Refresh those properly, meaning a substantial rewrite rather than a date bump. Everything else in the quarter goes to new posts, into clusters you don't cover.
Over a year on a weekly cadence, that's roughly 8–12 refresh slots out of 52 — near a fifth of your output. It converges on the 20–30% of slots we suggested reserving for updates in our look at how often a small blog should actually publish, with one difference: here the number is derived rather than decreed, which matters because it moves. Once a dozen posts have measurable history, the rule surfaces more candidates and the refresh share rises on its own. A 300-post blog genuinely should refresh more than a 30-post one — not because refreshing is inherently better, but because it has more pages capable of decaying.
Orbit Media is the cleanest small-team proof that this endpoint is reachable: roughly half its library is now rewrites, 50+ posts have been updated once and 15+ three or more times, and its survey work found that bloggers who update old articles are twice as likely to report strong results. Andy Crestodina's phrasing — "You'll get better results with less effort" — is the reward at the far end of a library you have to build first.
Four honest objections, because you'd raise them anyway.
Publishing frequency correlates with results too. Orbit Media's 2025 survey of 808 content marketers found 37% of those publishing multiple times per week reported strong results against a 21% overall benchmark, and 39% of those writing 2,000+ word posts did. Both camps can point to correlations in the same dataset. Neither is causal, and the fair reading is that both correlate with effort — of which you have a fixed amount.
Refresh case studies are survivorship-biased. HubSpot's 106%, individual blogs reporting 61.9% organic growth over six months — self-reported by parties with an interest in the result, none with control groups. The two studies here that did use controls report smaller, more conditional effects, and Workshop Digital states outright: "Not every refresh worked. Some updated pages underperformed after launch."
The zero-click decline may be stabilising. Seer's data shows AI Overview CTR bottoming at 1.3% in December 2025 and recovering to 2.4% by February 2026, with non-AIO searches improving from 2.8% to 3.8%. Any strategy built on zero-click as a permanent one-way ratchet will look wrong within a year.
Nobody has studied blogs this small. Every dataset here is a large blog, an agency client base, or a random-URL sample. There is no published refresh-versus-publish study on blogs under 50 posts. This post reasons from mechanism — power law, ranking odds, citation recency — and you should hold it that loosely.
The quarterly loop survives being wrong about any one of them. That's the point of preferring a rule to a ratio: when the evidence shifts, you change a threshold, not a strategy.
How often should I update old blog posts on a small blog? Audit quarterly, then act on what the audit surfaces — typically two or three refreshes per quarter on a 30-post blog. Kevin Indig calls refreshing anything older than 12 months a "no-regret move," but on a small library most posts won't have enough traffic history to justify the work. Check the signals rather than the calendar.
Does changing the publish date help SEO? No, not on its own. Google's documentation says dates should reflect genuine publication or significant update, and that Google estimates the date itself rather than trusting your declaration. John Mueller has confirmed that date changes without substantive edits don't improve rankings. Fixing typos or swapping an image doesn't warrant a new date.
How much do I need to change for it to count as a refresh? In the only controlled study available, only pages where 31–100% of the content changed showed a statistically significant ranking gain (+5.45 average positions). Edits under 30% performed in the same range as never-updated control pages. Plan for a rewrite of the substance, not an appended section.
Should I delete posts that never got traffic? Usually not first. Google's core-updates guidance calls deleting content "a last resort, and only to be considered if you think the content can't be salvaged." Consolidating two thin posts into one strong page with a 301 is generally better; outright pruning is for very short, zero-traffic pages that are off-topic or about obsolete tools.
My clicks dropped but my rankings held. Should I refresh? Check impressions. If impressions held or rose while clicks fell, that's zero-click capture — an AI Overview or featured snippet answering the query on the results page — and a rewrite won't recover those clicks. The productive response is making the page more citable rather than longer.
The allocation question answers itself once you stop treating it as a ratio. Every quarter, check which of your handful of ranking posts has lost about half its impressions while the SERP still wants what you wrote, and rewrite those properly. Everything else goes to new posts, in clusters you don't yet cover. On a 30-post blog that lands near one refresh in five slots — and rises naturally as your library earns more history worth defending.
The hard part isn't the rule; it's holding a cadence long enough for the rule to have something to work on. That's the job Magic Share was built for: an agent that researches your niche, drafts a verified post on your schedule and holds it for approval, so a steady cadence doesn't depend on you having a free Tuesday. Your first three posts are free — enough to see whether a drafted post is one you'd actually publish.
Magic Share researches, writes and fact-checks posts like this for any site — point it at your URL and review your first draft today.
Plant your first post