Every page you publish starts losing value the day it goes live.
Publishing more pages just means more pages losing value. The programs still gaining in year three are the ones that budget for keeping pages current.
What it costs to keep one page current.
A page will never be more accurate or more relevant than the day it goes live. Everything around it keeps moving. People start searching for something slightly different, competitors publish, and the facts inside the page quietly age.
The analogy we use is a car. Drive one off the lot and it starts losing value, because the technology in it is as good as it will ever get. A Tesla gets new capability by software update, so a car you already own gets better. Most content programs are built like the first car.
Every page carries what we call service debt: the person, effort, time and energy it takes to keep one page accurate, compliant and ranking. It recurs, and it sits on nobody’s budget line. A program that ships 30 pages a quarter and refreshes none of them adds 120 pages a year to a maintenance list nobody budgeted for, against a team that is not growing.
Somebody measured how fast that debt comes due. SE Ranking published 2,000 unedited AI-written articles across 20 brand-new domains in November 2024 and tracked them for 16 months. 70.95% indexed within 36 days; indexing was never the hard part. The share reaching the top 100 fell from 28% in month one to 3% by month three. And over the 16 months the set produced 1,092,079 impressions against 1,381 clicks, which is what positions 40–90 look like. The impressions keep arriving and the clicks never do.
One experiment is easy to wave off, and the same shape turns up in the field. Lily Ray monitored more than 220 websites publicly identified as customers of AI content-scaling platforms. 54% lost 30% or more of their peak organic traffic, 39% lost half or more, and 22% lost three quarters or more.
Google’s policy is aimed at the purpose behind the pages, not the tools that made them. Its spam policies call scaled content abuse “many pages generated for the primary purpose of manipulating search rankings and not helping users.” After the March 2024 core update, Google reported 45% less low-quality, unoriginal content in its results, against a 40% target. We use AI tooling heavily, with a person reviewing what ships and a plan for maintaining it. The problem is volume with nobody behind it. Purely AI-generated, unedited pages are the ones that lose: Semrush’s 42,000-page analysis found pages its GPTZero classification called purely AI-generated held position 1 10% of the time, versus 80.5% for human-written pages, and Semrush notes that AI-detection classification is imprecise.
The other half of this is that the top of Google’s results is old. Across 1 million URLs and 1.3 million keywords, Ahrefs found only 1.74% of newly published pages reach the top 10 within a year, 72.9% of current top-10 pages are more than three years old, and the average #1-ranking page is five years old. Position accrues to history, and volume buys no shortcut through it.
A refresh has to be substantive to count. Google’s own self-assessment questions are the test we would hand a team: “Are you changing the date of pages to make them seem fresh when the content has not substantially changed?” and “Are you producing lots of content on many different topics in hopes that some of it might perform well in search results?”
Both are checks a program can run on itself today.
Ranking is one problem, and staying there is a different one that costs more than most people expect.
Going from challenger to incumbent changes the logic, and at the top it is tempting to ease off. We think that is the trap. The people chasing you never stop, and holding a position takes the same system that earned it.
The same goes for AI summaries. Pew Research Center found that with an AI summary present, users clicked a traditional result on 8% of visits versus 15% without one, and clicked a link inside the summary on 1% (Pew Research Center, Chapekis & Lieb, 2025). Semrush found commercial-intent results carrying an AI Overview grew 71% between November 2025 and April 2026 (Semrush, Luke Harsel, 2026). Ahrefs found only 12% of AI-cited URLs rank in Google’s top 10 for the original prompt (Ahrefs, Linehan & Guan, 2025), so ranking well helps you get cited, and it doesn’t guarantee it.
Six checks you can run on your own site this afternoon.
All six take a CMS export and Search Console. You do not need a vendor, us included, to run any of them.
Nothing substantively edited since it shipped
Export every URL with its last-modified date, then compare pages published per quarter against pages updated per quarter. If publishing has outrun updating every quarter since launch, the liability is growing faster than the asset.
The refresh was a date change
Diff five pages carrying a recent “Updated” stamp against the version before them. If what changed is the date, the year in the H1 and one swapped statistic, that is the pattern Google’s own quality documentation names.
Impressions climbing, clicks flat
Plot 16 months of impressions against clicks for whichever folder grew fastest. Lines that come apart are the signature of a page set sitting at positions 40–90, seen and never chosen. The SE Ranking experiment is the extreme version: 1,092,079 impressions, 1,381 clicks.
The traffic is an inheritance
List the top 20 pages by organic sessions with the publish date and the last meaningful edit next to each. If the median publish date is three or more years back and nothing has been reworked in 18 months, the program is living on pages a previous team wrote.
The cluster doesn’t link to the page it supports
Take the commercial page a cluster was built to support and count the internal links pointing at it from inside that cluster. If the supporting pieces link to the blog index and to each other but never to the page they exist to feed, the cluster is decoration.
Nothing has ever been retired
Count the URLs with zero clicks in the last 12 months, then work out what has ever been merged, redirected or removed. If the answer is nothing, all of them are still on the maintenance list, still holding internal links that would do more work elsewhere.
What a maintained portfolio looks like two years in.
Sherpa ran two years of continuous publishing and maintenance. Here is what that did.
~8,500 → ~79,000 monthly organic sessions
June 2024 → June 2026, roughly 9× over two years of continuous publishing and maintenance.
“Last spring our organic traffic was ballooning at such a rate, it made a massive shift in our marketing strategy easy… If we hadn’t seen that growth, we’d still be paying affiliates tens of thousands a month for low-performing leads.”Scott Klemm · Sherpa Auto Transport
You can watch the loop run in a partner’s own numbers: good content earns links, links earn traffic, traffic earns reviews and margin, and margin pays for more content. Year two is where it showed: we kept improving the system and the curve got steeper. The figures are third-party estimates, and content was one part of a program that also carried technical work and design.
None of this means waiting a year. Franzy saw roughly a 20× organic lift on a stalled directory folder within six months of relaunch, with about 30 days to a working engine and first results about two months in.
Source: A2 Analytics — Franzy case study.Seven steps, in the order we’d do them.
The ranges below are engagement ranges for a program of roughly 150–5,000 URLs. They describe how long a step tends to run, not when results turn up.
Build one inventory of what already exists
Join the CMS export, a full crawl, Search Console, analytics and the sitemaps together on a single URL key. Most of the time goes to cleanup: normalizing trailing slashes and case, reconciling timezones, and working out which of four near-identical URLs people actually land on. One to three weeks.
Score decay, put a disposition on every page
Keep and maintain, refresh, merge into a stronger page, or retire. In practice this is a series of arguments with product, sales and legal about what a page is for, and no tool produces it for you. Expect to find page types nobody currently owns, and expect at least one of those arguments to end with somebody going off to ask a person who left the company.
Do the kill list before publishing anything new
Merges, redirects and removals all happen before a single new page gets written. It feels like going backwards, and it is the step partners skip most often, usually because retiring pages is a hard thing to put in a quarterly deck. One to two weeks of decisions, and longer if legal has to look at anything that was ever a claim.
Wire the first-party data in
Pricing, reviews, availability and operational tables have to come out of production systems and into a shape a page can render. This is a ticket sitting in somebody else’s engineering backlog, and it is the single most underestimated part of the work. It is also where a large services company with a lot of transactional data has an advantage nobody can buy. Weeks, and most of the elapsed time is waiting.
Stand up the production pipeline end to end
Research, drafting, editing, compliance review, design, publish, and then the same run again the following week. In regulated verticals the compliance gate is the slowest thing in the whole program, and it does not get faster just because the drafting got faster. Push a handful of pieces all the way through before you scale any of it, because that is where you find out who the real reviewer is.
Set the refresh triggers
Some triggers run on a calendar and some run on signals: a ranking slipping past a threshold, the shape of Google’s results changing under you, a source table going stale, a price or product change upstream. The point of writing them all down is that maintenance stops being something a person has to remember on a Friday afternoon.
Watch the portfolio curve, not single pages
Individual pages move unevenly, and Ahrefs found 40.82% of the pages that reach the top 10 get there within a month while only 1.74% of new pages get there within a year at all (Ahrefs, Patrick Stox, 2025). At the portfolio level the realistic range is two to four quarters before the shape of the curve is readable.
Two ways next year’s budget can go.
Measured in pieces per month
There is a calendar, a quarterly deck, and a library that grows on schedule, and running it feels productive. Traffic climbs for three or four quarters and then flattens. Because the metric everyone watches is production, the answer is more pieces.
Twelve months in there is a folder of pages nobody remembers commissioning, and the team is busier than it has ever been. The pages actually carrying traffic are the oldest ones on the site, and nobody has opened them since the day they shipped.
Measured in what the pages earn
The budget gets split before the year starts, some to new pages and some to maintenance, written down where everyone can see it. The first deliverable is a count rather than a calendar: how many URLs exist, how many earn anything, and how many are still current.
Quarters one and two look slower if you are still measuring output, and sitting through that is the price of admission. By year two the same budget produces more, because the work lands on pages that already have history behind them.
of the pages in Google’s top 10 are more than three years old.
Source: Ahrefs, Patrick Stox, “How Long Does It Take to Rank in Google? And How Old Are Top Ranking Pages?,” 2025 · 1.3M keywordsWhere these numbers come from.
Share of pages in the top 100, month one to month three, across 2,000 unedited AI articles
Source: SE Ranking, “How AI-Generated Content Performs: Experiment Results,” 2026Clicks against 1,092,079 impressions over 16 months, same experiment
Source: Search Engine Land, “How AI-generated content performs in Google Search: A 16-month experiment,” 2026Of newly published pages reach the top 10 within a year
Source: Ahrefs, “How Long Does It Take to Rank in Google?,” 2025Of 220+ AI-scaled sites lost 30% or more of peak organic traffic
Source: Search Engine Journal, Lily Ray, “It Works Until It Doesn’t,” 2026Position-1 share, human-written vs AI-classified pages across 42,000 pages. AI-detection classification is imprecise, per the study
Source: Semrush, “Does AI content rank well in search?,” 2026Click rate on traditional results with vs without an AI summary present
Source: Pew Research Center, “Google users are less likely to click…,” 2025Less low-quality, unoriginal content in results after the March 2024 core update, against a 40% target
Source: Google, Elizabeth Tucker, The Keyword, 2024Start with a count of what you already have.
We will go through your portfolio and tell you how many URLs exist, how many earn anything, and how many are still current. Then where we think the split between new pages and maintenance should land.