Every B2B content programme we inherit has the same shape. Four to eight years of publishing, somewhere between 300 and 1,200 indexable URLs, and a distribution where roughly the top 8% of pages earn nearly all of the non-branded traffic. The remaining 92% are not neutral. They are competing with the pages you care about for internal links, for crawl budget, and — the expensive one — for the site’s topical identity.
The instinct is always to publish more. The correct move is almost always to publish less and delete a lot.
Why volume stopped paying
Three things changed, and they compounded.
Crawl allocation follows demonstrated value. A site where most URLs have not been requested by a user in nine months gets crawled less, and re-crawled more slowly, than a site of a third the size with the same demand. The new page you just shipped is behind two hundred pages of 2021 listicles in the queue.
Site-level quality assessment is real. You do not need to believe in a mystical “site quality score” to accept the mechanical version: if a large fraction of your pages fail to satisfy the queries they rank for, the aggregate signal your domain sends on any new page is worse. We have watched a single pruning pass move rankings on pages we never touched.
Retrieval punishes near-duplication hard. Six posts that each half-answer the same question produce six mediocre chunks that dilute one another in embedding space. One post that fully answers it produces one strong chunk. Answer engines do not average your six attempts; they pick somebody else’s one.
The three questions that decide every page
Do not start from traffic. Start from role. For each URL, in order:
- Does anyone arrive here on purpose? Entrances from search or referral in the last twelve months, not pageviews. A page that is only ever reached by clicking through from another page of yours has no independent demand.
- Does it do a job nothing else does? Not “is it about a unique keyword” — is there a question a buyer asks that only this page answers?
- Would you show it to a prospect? If the honest answer is no, it is not neutral. It is a page a prospect might find.
Four numbers make those answerable: organic entrances (12 months), assisted conversions or pipeline touches, referring domains, and internal links in. Pull them into one sheet before you make a single decision. The argument about any individual page always dissolves once the four numbers are next to each other.
The decision table
| Signal | Keep | Merge | Retire |
|---|---|---|---|
| Organic entrances (12mo) | Meaningful and stable | Thin but non-zero | Effectively none |
| Unique job on the site | Yes | Overlaps a stronger page | None |
| Referring domains | Any | Any | None |
| Would you send it to a buyer | Yes | After a rewrite | No |
| Action | Refresh on a schedule | 301 into the survivor | 410, or 301 if it has links |
Two notes on the action column, because this is where implementations go wrong.
Use 410, not 404, for deliberate removal. Both remove the page. 410 Gone
states that the removal was intentional, and in practice it drops out of the
index faster and stops being re-requested sooner. A 404 leaves ambiguity you have
no reason to leave.
Do not 301 everything to the homepage. A redirect to an irrelevant target is treated as a soft 404 — you get the drop without the signal consolidation, plus a redirect chain to maintain forever. If there is no relevant survivor, the page should 410. Redirect only where the destination genuinely answers the query the old page ranked for.
What “merge” actually means
Merging is not concatenation. Four thin posts stapled together make one long thin post, and you have converted four weak pages into one weak page with worse scannability.
The version that works:
- Pick the survivor on link equity and URL sanity, not on which draft you like.
- Take from each donor only the passages that state something the survivor does not. Usually that is one or two paragraphs out of eight hundred words.
- Rewrite the survivor’s structure around the combined scope, including new section headings. If the merged page has the old page’s outline, you merged the text and not the argument.
- 301 the donors. Then fix the internal links that pointed at them — do not leave the site’s own navigation depending on redirects, which is how a redirect chain becomes four hops in two years.
What you will get wrong
The two failure modes, both of which we have caused:
Pruning the long tail that was actually working. Aggregate reports hide pages that produce two qualified demos a year at zero marginal cost. Set a floor — any page with a conversion or a referring domain survives the first pass by default — and review the exceptions by hand.
Pruning in one release. Retire in batches of 50 to 100 URLs, two to four weeks apart, with the batch recorded. When something moves you want to know which batch moved it. A single 400-URL release gives you one data point and no way to attribute it.
Measuring it honestly
Track four things, and be patient with all of them: indexed URL count against submitted, non-branded impressions on the surviving pages, crawl requests per day from the log files, and pipeline attributed to organic. Expect sessions to fall for four to eight weeks. That dip is the pruned pages doing exactly what you predicted they would stop doing.
The programme that fails is the one where somebody looks at the dip in week three, panics, and restores 200 URLs. Agree the measurement window before you delete anything, in writing, with whoever will be looking at the dashboard.