We had been publishing three or four pieces a week for two years, on the theory that volume compounds. Traffic had been flat for eight months and we could not work out why.
We stopped publishing entirely for four months and spent that time on the existing library. Organic traffic rose by roughly a third over that period and kept rising afterwards.
What the library actually looked like
Auditing it was the uncomfortable part. About four hundred pieces, of which:
Roughly a quarter received essentially no traffic and never had.
Another quarter received a trickle and were substantially similar to each other, covering overlapping topics with different phrasing.
About a fifth were outdated in ways that made them actively wrong.
And a small proportion, perhaps thirty pieces, accounted for the overwhelming majority of the traffic.
That distribution is apparently normal and it was not what any of us expected.
What we did with each category
The zero-traffic pieces we mostly removed, redirecting to the nearest relevant page. About eighty pieces went.
The overlapping ones we consolidated. Where five thin pieces covered aspects of one topic, we merged them into one substantial piece and redirected the rest. That turned about ninety pieces into eighteen.
The outdated ones we updated properly — checking every claim, rewriting sections, changing examples — rather than editing the date and adding a paragraph.
And the small group that performed well we left almost entirely alone, except to improve internal links into them, on the principle of not interfering with what works.
Why it worked, as best we can tell
We cannot prove causation and the pattern is consistent with what a lot of practitioners report.
Consolidation seems to help because five thin pieces on one topic compete with each other and none is comprehensive. One substantial piece is a better answer and accumulates the signals that were split five ways.
Removal seems to help because a large body of pages nobody engages with may affect how the whole site is assessed. Whether this is a formal mechanism or an emergent effect is debated and the practical result is widely observed.
And updating helps straightforwardly, because content that is accurate is more useful than content that is not.
The part that was hardest
Deleting things. There was genuine resistance, and the arguments against were the ones you would expect.
Somebody spent time writing it. It might get traffic eventually. It does not cost anything to leave it there.
The last one is the argument that needed answering, and the answer is that it does cost something — it dilutes the topical focus of the site, it splits internal linking, and it means anybody assessing the domain encounters a lot of material that is not good.
What made it possible was checking traffic and links for each page before removal, and redirecting rather than deleting outright, which preserved anything of value.
What we changed about publishing afterwards
We publish about a third of what we used to.
Each piece takes considerably longer and covers a topic properly rather than partially.
Before writing anything new, we check whether an existing piece should be expanded instead. That check kills perhaps half of proposed articles, and improving the existing one is nearly always the better outcome.
And we schedule updates to existing content as work in its own right, with time allocated, rather than treating it as something to do when there is a gap.
The caveats
Being honest about what this does not prove.
We changed several things at once, so we cannot separate the effects of removal, consolidation and updating.
The four month period included changes in the wider environment that we cannot control for.
And our starting position was a large body of poor content, which is exactly the situation where this approach helps most. A site with a small library of good material would gain nothing from a pruning exercise.
The generalisable part is the audit itself. Establishing what proportion of your library does anything at all takes an afternoon, and for most sites that have been publishing at volume the answer is uncomfortable and actionable.
How we ran the audit
Since the method matters and it is not complicated.
Export every URL with organic sessions, impressions and clicks for the last twelve months, plus any external links pointing to it.
Sort by sessions and look at where the distribution falls off, which is usually dramatic.
Then categorise manually: keep, update, merge, remove. That manual pass is the slow part and it cannot be automated sensibly, because deciding that five pages should be one requires reading them.
Four hundred pages took two people about a week, which is a reasonable cost for a decision that shaped the following year.
The redirect discipline
Everything removed was redirected to the most relevant surviving page rather than to the homepage, which preserves whatever external links existed and avoids sending readers somewhere useless.