SaaS content refresh

Every content team knows it should update old posts. Almost none can tell you which posts, in what order, or what the last round of updates actually returned. The work gets done in bursts, usually in January, usually on whichever posts someone remembers, and the results get reported as a total traffic number that seasonality explains just as well.

An operating model fixes that. Four decisions, one detection query, and a control group.

Detect decay with one Search Console comparison

Open Search Console, go to Search Results, set the date range to the last 90 days, then use the compare tab and select the same 90-day window from the previous year. Switch to the Pages tab and export. That single comparison is the whole detection layer, and it beats month-over-month reporting because it cancels out seasonality, which is brutal in B2B software where July and December collapse every year regardless of what you published.

Three columns matter: clicks, impressions and average position. What separates a real problem from a structural one is the relationship between them.

That first row is the one the category keeps misdiagnosing. AI Overviews now appear on roughly half of all queries and a much higher share of B2B technology queries, and the reported click impact ranges from about 15 percent to 61 percent depending on which study you read. The studies disagree because they measure different query sets with different methods, which is a good reason to measure your own site rather than quoting any of them.

Three filters make the export usable. Drop any URL below 20 clicks in the prior-year window, because the percentage swings on small numbers are noise. Drop anything published in the last 6 months. And segment the remainder by page type, since a comparison page losing 30 percent means something very different from a top-of-funnel explainer losing 30 percent. The definitions behind all this sit in our content decay entry if you need a shared vocabulary for the team.

For sites past a few hundred pages, move this to BigQuery via the Search Console bulk export and schedule it monthly into Looker Studio. The manual export stops being practical somewhere around 400 URLs.

Run every flagged page through one decision tree

Four outcomes. No fifth option, no ‘monitor it’, because ‘monitor it’ is how 200 decaying pages sit untouched for two years.

If the page...	Decision	Effort	Expected result
Ranks 4 to 20, structure matches the current SERP, facts are stale	Refresh	4 to 8 hours	Recovery within 4 to 8 weeks in most cases
Overlaps another URL on the same intent, or impressions split between two pages	Consolidate and redirect	6 to 12 hours	Combined page usually outranks both within a quarter
Under 10 clicks in 12 months, no links, no conversions, off-topic for the product	Prune	30 minutes	Cleaner crawl budget, no traffic loss
Down but the dip matched last year, or impressions and position are flat	Leave alone	0	Recovers on its own or is structural

The consolidation branch is the most valuable and the most avoided. Nobody wants to delete a post that took a week to write. But two pages targeting the same intent split impressions, split links and confuse the engine about which one to serve, which is textbook keyword cannibalisation. Merging the stronger content into the better-performing URL and 301-redirecting the other reliably outperforms improving either one alone.

Pruning deserves a stated threshold so it stops being a judgement call. Mine: under 10 organic clicks in the trailing 12 months, zero referring domains, zero form completions, and no topical relationship to what the product does. A page hitting all four gets a 301 to the nearest relevant page if it has any inbound links at all, and a 410 if it has none. Redirecting everything to the homepage is not pruning, it is hiding the problem.

One constraint worth writing down before you start: never prune or consolidate during a migration or a domain change. If you are still deciding where the blog lives, settle that first, because the subfolder versus subdomain decision changes which URLs you are consolidating into.

What to do when the loss is structural rather than a quality problem

Structural loss is the case the older refresh advice has no answer for, and it is now the most common pattern on informational pages. Impressions hold, average position holds, clicks fall. The page still ranks. The click is being spent above it.

You have four responses, and rewriting is not one of them.

• Consolidate. Three thin informational posts on adjacent questions are worth more as one deep page that an engine will cite than as three pages that each lose their click. Merge and redirect.
• Convert the page’s job. If the post can no longer earn a session, make it earn an internal link instead. A well-linked explainer that passes authority to a converting page is doing real work even at a quarter of its old traffic.
• Make it extractable. Add a 40 to 60 word answer capsule under each H2, one table, and a dated figure with a named source. If the click is going to an answer, the goal is to be the source that answer names.
• Retire the traffic target. Change the reported metric on that page from sessions to citations and assisted conversions, and tell whoever reads the dashboard that you changed it and why.

The refresh checklist itself

A refresh is not ‘add 300 words and change the date’. Google’s own guidance warns against changing dates to signal freshness without substantive change, and a date bump with no real edit gets discounted. Work the list.

That last item is the one people skip and then regret, because without a baseline recorded before the edit there is no way to claim a result afterwards. Capture it in the same sheet you use for your keyword map so the target term and the baseline live together.

Roughly a third of refreshes will not move anything. That is the honest base rate, and the usual cause is that the page was never going to rank for that term because the domain lacks the authority or the SERP is owned by review sites. A refresh cannot fix a targeting error, and a team that refuses to accept that will refresh the same page three times.

Cadence by page type, not by calendar

A site-wide ‘update everything annually’ rule wastes effort on pages that never change and neglects pages that go stale in weeks.

The last row changes how you think about scale. Programmatic pages are refreshed by updating the template and the underlying data, which means a single change touches thousands of URLs. That is a very different operating cost from editorial refresh, and it is one of the practical differences we set out in programmatic versus editorial content. If your refresh involves changing a page pattern rather than a page, write it as a spec using the programmatic page brief template.

Prove the lift with a held-back control group

Most refresh reporting is worthless because it compares refreshed pages to their own past, during a period when the whole site was moving anyway. If organic is up 12 percent site-wide, refreshed pages being up 14 percent is nearly nothing.

Do this instead. Take the 40 pages your detection query flagged. Sort them by traffic band and page type, then assign every other page to a control group so the two sets are matched, giving you 25 to refresh and 15 to hold. Refresh the treatment group in one batch over two weeks. Leave the control completely alone for 90 days.

A realistic outcome for a first cycle on a library of 150 to 400 pages: the treatment group median gains somewhere in the range of 15 to 60 percent more clicks than control over 90 days, about a third of individual pages show no movement, and one or two pages drop because the rewrite chased the wrong intent. Report all three. A refresh program that only reports winners loses credibility the first time someone checks.

Budget: how much of content spend belongs to refresh

Below roughly 150 published pages, almost all of it should go to net-new. You are still building coverage and there is not enough decay to chase.

Past 150 pages, my position is that 30 to 40 percent of content spend should go to refresh, consolidation and pruning, and that pruning should appear as a named quarterly output with a page count attached. Teams resist this because publishing feels like progress and deleting feels like failure. The arithmetic disagrees: refreshing a page that ranks 8th and recovering it to 3rd is usually cheaper per incremental session than commissioning a new article that may never rank at all, and you can check that claim for your own numbers with the SaaS SEO ROI calculator or model the ranking upside with the organic traffic forecast calculator.

Above 500 pages the split often inverts, and the best-run content teams I know are spending more on maintenance than on new commissions. That feels wrong to a CMO who bought a content program to produce content. It is what the data says.

What to do this week

Pull the year-over-year Search Console comparison today. It takes fifteen minutes and it will tell you something uncomfortable, usually that between a quarter and a half of your library is losing clicks and nobody noticed.

Then classify the top 40 decayers against the decision tree, split them into treatment and control, and put the first refresh batch on the calendar as a two-week sprint rather than a background task. The full operating cadence, including who owns classification and how to run this monthly without it eating the roadmap, is in the content refresh program playbook, and where refresh sits against technical and link investment is covered in the SaaS SEO pillar.