TL;DR: Wikipedia is the most-cited corpus inside every major LLM and answer engine. Most growth teams treat it as a no-go zone because of its editor immune system. We found the door: shape your owned content as a citable secondary source, not a brand asset. 47 of 312 source candidates survived 90 days of editor scrutiny, and downstream AIO / Perplexity mentions of those URLs lifted 9.0x.
The hypothesis
Wikipedia's castle is guarded by volunteers, bots and a 20-year-old policy manual. The front gate (paid placement, brand-page edits, ref-spam) is welded shut. But the castle has a service entrance: articles that need better sources. Every "citation needed" tag is an open door for a reference that meets WP:RS.
If we publish independent, methodology-first research that fills a real factual gap, then editors will place it themselves — and the citation will compound into every answer engine that trains on Wikipedia.
The setup
- Article pool. 1,140 Wikipedia articles in our domain space with at least one open {{Citation needed}} tag, sorted by monthly pageviews.
- Source candidates. 312 long-form research pages we published, each filling exactly one factual gap with original data, transparent methodology and a third-party author byline.
- Door. Edits proposed through Talk pages and WP:RFC, never direct refbombing. Half the placements made by independent editors after the source was indexed in Google Scholar.
- Metric. 90-day citation survival, plus downstream cited-mention rate in AIO, Perplexity, ChatGPT Search, Claude, and Gemini.
What we saw
Of 312 source candidates, 64 were placed within 30 days and 47 survived the full 90-day window. The reverted 17 fell to one of three failures: promotional tone, single-source claim, or an editor flagging the byline as affiliated.
The downstream signal was the headline. Pages cited by even one Wikipedia article were quoted by AI Overviews 9.0x more often than identical pages that never got placed — and Perplexity citations rose 7.3x. Wikipedia is the trust spine the answer engines lean on.
The signals that survive
We isolated which source-level signals correlated with surviving 90 days. Four signals explained 83% of the variance.
What ships next
We are packaging the source-archetype playbook inside Agentic GEO Engine and rolling it across a 2,400-article candidate pool across nine non-English Wikipedias, with parallel tracking on every major answer engine.
