GUEST POST · CONTENT Aug 16, 2026 · 6 min read

The pruning trade: what deleting 40% of your pages does to AI citations

JT June Tanaka Head of Search, Fintech · 6:00 AM ET 𝕏 in ✉

Editorial illustration: a dense block of pages dissolving to leave a few solid survivors

Ten years of content marketing produced a specific kind of wreckage: thousands of thin posts written for keywords nobody searches, published because the calendar said Tuesday. In classic SEO that archive was mostly harmless dead weight. In AI search it is worse than harmless, and the published data on pruning is strong enough to act on.

⚡TL;DRDomains that pruned more than 40% of their pages saw AI citation growth rise by an average of 62% in one 2026 analysis. The mechanism is entity clarity and crawl concentration, not page count. But prune with a citation audit in hand, because engines keep citing pages after you delete them, and the pages already earning citations are the ones most likely to look prunable on traffic alone.

Why the archive actively hurts

A large thin archive does three things to an answer engine. It splits your entity signal across dozens of near-duplicate pages so nothing reads as the canonical answer. It spends crawl budget on pages nobody will ever cite. And it puts your own pages in competition with each other for the same passage retrieval. The 2026 pruning analysis found domains cutting more than 40% of their pages recorded average AI citation growth 62% higher than those that did not.

AI citation growth after aggressive pruningPruned 40%+ of pages+62% citation growthNo significant pruningbaselineAverage AI citation growth for domains pruning more than 40% of pages. Source: Searchless content pruning analysis, 2026. Directional: vendor research on a self-selected sample.

The trap: engines cite pages you already deleted

Here is the finding that should slow anyone down before a mass delete. Otterly's dead citations experiment found AI engines continuing to cite pages after removal, sometimes long after. A deleted page can keep appearing as a source in answers, which means the experience it delivers is a 404 with your brand attached to it. Pruning is not instant in the retrieval layer even when it is instant on your server.

"A deleted page can keep showing up as a source. Pruning is instant on your server and slow in the retrieval layer, and the gap is a 404 with your name on it."

The practical implication is that redirects matter more than deletions, and that you should never prune from a traffic report alone. Pages earning AI citations frequently show poor classic traffic, exactly the profile that gets cut in a conventional content audit.

A method that does not destroy assets

THE PRUNING SEQUENCE011. Inventory citation assets firstBefore touching anything, list every URL currently earning AI citations. These are retrieval assets regardless of what their traffic looks like. They do not get deleted, and they rarely get redirected.022. Sort the rest into update, merge, deleteUpdate anything with genuine potential. Merge overlapping pages into one canonical answer with a 301. Delete only true dead weight: no traffic, no links, no citations, no purpose.033. Consolidate rather than remove where possibleMerging three thin posts into one strong page keeps the equity and creates the canonical passage. Deletion is a last resort, not the default.044. Return the right status and clean the linksUse 410 for genuine removals so the signal is unambiguous, and strip internal links pointing at removed pages so you are not advertising your own dead ends.055. Refresh what survives, substantivelyRecency is a measured citation factor, and it requires real change. New data, new sections, corrected facts. A touched date with untouched content does not earn it.

That last point has its own evidence base. Seer Interactive's recency study found substantively updated content earning materially higher AI citation rates, while date-only edits did not carry the effect. Pruning and refreshing are the same project: fewer pages, each one genuinely maintained.

What to do this week

Pull your citation report before your traffic report. Mark every URL earning citations as untouchable. Then take the bottom 20% of what is left by every measure at once, no traffic, no links, no citations, no strategic purpose, and merge or remove it properly with redirects and clean internal links. Do not do four thousand pages in a week. Do the worst two hundred, wait a month, and watch whether citation share on the survivors moves. It is a slower answer than the headline number suggests, and it is the one that does not cost you assets you did not know you had.

KEY TAKEAWAYS01Domains pruning 40%+ of pages saw average AI citation growth 62% higher in the 2026 analysis. Treat as directional vendor research.02The mechanism is entity clarity and crawl concentration, not page count itself.03Engines keep citing deleted pages. Prefer merges and redirects over deletion, and expect lag in the retrieval layer.04Never prune on traffic alone. Citation-earning pages often look weak on classic metrics.05Refresh substantively. Date-stamp edits without real change do not earn the freshness premium.