Last month I screened a website with 412 pages. In Search Console, 63 of them had together delivered zero clicks and zero impressions over twelve months. The content calendar for the next quarter was packed anyway: twenty new blog posts. When I asked why, the answer was: "That's just what you're supposed to do."
I see that pattern in seven out of ten audits. There's a visibility problem, so the solution looks like more content. In classic SEO that logic mostly held up: more pages means more surface area, more long-tail, more chances. In GEO — where ChatGPT, Perplexity and AI Overviews assess your site passage by passage — it works against you. Every weak page dilutes the signal your strongest pages send.
Content pruning isn't a tidy-up job for a quiet Friday afternoon. It's a visibility strategy: you remove what thins out your credibility and keep what a language model can attribute to you without doubt.
Why volume weakens your GEO visibility
A language model reads passages, not sitemaps
ChatGPT doesn't fetch a website. It fetches a piece of text — a chunk — compares it against what other sources say, and formulates an answer from that. In that process your site isn't a whole; it's a collection of loose pieces that may or may not get picked up.
Say you have three articles about "GEO for e-commerce stores": one from 2021 with outdated tips, one from 2023 with a slightly different definition, and a FAQ page from 2025 claiming something else again. A retrieval system grabs the passage that fits the question best. That's a gamble, and the gamble rarely lands in your favour. If the model finds two contradictory passages from the same source, the reliability of that source as a whole drops. You're competing with yourself for the same chunk.
Thin pages weaken your entity
Authority in AI systems hangs on entities, not URLs. An entity is a thing with a recognisable identity: your company, your specialism, your approach. You don't build that with forty half-finished articles, but with a handful of deep sources that use the same concepts, definitions and connections consistently. That's exactly what entity SEO is about — and it's why a competitor with 80 pages gets cited more often than you with 400.
Bloat costs you crawl budget, speed and maintenance
Thin pages cost you three times over. AI crawlers waste their budget on pages without value and reach your good work less often. Paginated archives, tag pages and outdated news posts drag down your Core Web Vitals — which AI crawlers watch just as closely as classic search engines, as I describe in Core Web Vitals and AI crawlers. And every quarter you maintain content nobody has ever read.
The Prune Matrix: four actions, one decision per page
Go through your pages one by one and force yourself into one of these four.
| Action | Signal | What you do | |---|---|---| | Keep | Clicks, citations, backlinks, a unique angle | Leave it, just check the facts and the date | | Improve | Impressions but low CTR, outdated info | Rewrite, go deeper, add sources | | Merge | Two or more pages with the same search intent | Combine into one strong page, 301 the rest | | Remove | Zero clicks, zero impressions, no backlinks, no AI referrals | 410, or 301 to the closest alternative |
Keep: the quietest box
Pages that pull visitors in from ChatGPT or Perplexity always stay — even if they perform modestly in Google. AI referrals are a leading indicator, not a side note. Check your analytics for referrals from chatgpt.com, perplexity.ai and similar sources before you remove anything.
Improve: this is usually where the biggest win sits
Pages with impressions but a CTR below 0.5 percent aren't bad pages; they're badly phrased answers. Often there's no direct opening paragraph carrying the answer, or the point only lands in the final paragraph. Rewrite the first 100 words and the gain often shows up within two weeks.
Merge: same intent, multiple URLs
This is cannibalisation in its most common form. Three blog posts about "what does a website cost" with slightly different angles together deliver less than one page that genuinely explains it well. Merge them, build one page that covers everything, and set 301 redirects from the old URLs to the new one. That way the link value you've built up is preserved.
Remove: 410, 301 or noindex?
Rule of thumb: if there's a page that answers the same question better, use a 301. If there's nothing comparable, use a 410 — the page is gone and a crawler is allowed to know that. Noindex is only for pages that need to stay online but don't belong in search results, such as an internal manual. Never just drop removed URLs into robots.txt: that way a crawler never sees the 410 and the URL hangs around the system like a ghost.
A pruning sprint in six steps
- Gather your data. Search Console covering at least 16 months, analytics with AI referrals, a crawl with Screaming Frog, and your backlink profile.
- Segment. Zero clicks + zero impressions + zero backlinks = remove candidate. Impressions with a low CTR = improve. Overlapping slug or search intent = merge.
- Check AI referrals. Pages that bring in visits from ChatGPT or Perplexity stay. No discussion.
- Build the redirect map before you delete anything. Never the other way round. This is the step people skip, and where things go wrong.
- Update internal links and your sitemap in the same sprint. Removed URLs still sitting in menus or copy create dead ends.
- Monitor for 90 days. Clicks, impressions, and a fixed set of 30 to 50 prompts you use every month to test whether AI systems still mention you.
For the check before you publish or rewrite anything, I use the GEO checklist for 2026.
What it delivers, in numbers
For one B2B client we went from 380 to 196 pages. The first three weeks organic traffic dropped 6 percent — normal and temporary. After a quarter it sat 31 percent above the baseline, simply because each of the remaining pages picked up more internal links, more attention and more authority signals.
The GEO effect was more interesting still. In a monthly sample of 50 prompts about their niche, the number of times ChatGPT mentioned the company went from 2 to 11. Not because anything special had been added, but because there was nothing contradictory left to choose from.
That's the EEAT layer too: fewer pages means you can put a named author with a real role on each one, cite sources, share data and maintain a date. A page with a name, a job title and three verifiable sources beats five anonymous blog posts.
Three misconceptions about content pruning
"I'll lose traffic"
You lose pages, not automatically visitors. Pages without clicks deliver zero traffic; there's nothing to lose. For pages with traffic that you merge, that traffic moves with the redirect. Expect a dip of a few weeks, then recovery.
"Google penalises deletion"
Google penalises thin, undifferentiated content. Deleting is a solution, not a violation. What does hurt: mass 301-ing hundreds of URLs to your homepage. That makes the redirect unreliable and loses the context of the old page.
"You can't build authority without volume"
Authority is consensus between sources, not the number of URLs you own. Ten pages that contradict each other cost you more than three pages that tell the same story consistently.
Your next step
Set aside one afternoon. Export your URLs with clicks and impressions over twelve months, flag everything at zero, and check whether it overlaps with a page you already have. Don't delete anything until the redirect map is in place.
And if your team suggests just publishing "more content" — including with AI, which tends to enlarge the volume problem rather than solve it, as explained in content creation with generative AI — ask them which ten pages they'd remove this month. The answer tells you exactly how mature your content strategy is.
Frequently asked questions
How do I know which pages to remove?
Start with pages that have had zero clicks, zero impressions and zero backlinks over 12 to 16 months, and that also don't show up in your AI referrals. Those are almost always remove candidates. Borderline cases with a handful of impressions aren't deletions but improvements: rewrite them first and measure again after six weeks.
Will my rankings disappear if I delete a page?
The ranking of that specific URL disappears, but the topical relevance doesn't — as long as you redirect properly. Set a 301 to the page that answers the same question best, and update your internal links. Only use a 410 if there genuinely is no suitable destination. Expect a temporary dip of a few weeks.
Is content pruning useful for a small website too?
Especially then. With 30 pages, every weak page has a relatively large influence on how a language model positions your brand. For small sites, pruning is often a matter of merging three pages and deleting two outdated ones — an afternoon's work with a measurable effect on citations.
How often should you prune content?
Run a quick scan every six months: which pages got no clicks or impressions for six months, and which overlap in intent? Once a year, do a full audit with a redirect map and an updated prompt test. After a major repositioning or rebrand, do it straight away rather than on schedule.