Run enough prompts through the assistants and the same three citation names keep coming back: Wikipedia for what things are, Reddit for what people think, YouTube for how things are done. That concentration is not an accident, and it makes those three surfaces part of your visibility whether you participate or not.
Here is why models lean on them, what gets cited from each, and the ethical playbook per surface, with the shortcut version named for what it is.
Why these three
| Surface | Why models trust it | What gets cited |
|---|---|---|
| Wikipedia | The canonical reference layer of training data | Definitions, facts, what-is answers |
| Authentic user opinion at scale, licensed to AI labs | Recommendations, complaints, real-world takes | |
| YouTube | Transcribed how-to expertise, native to Google's stack | Tutorials, reviews, demonstrations |
Notice the division of labor: encyclopedia, forum, workshop. Between them they cover the intent spectrum, which is why they blanket the citations the way institutions blanket the health and legal categories.
The Wikipedia lane: accuracy first, ambition second
The blunt part: most businesses do not meet notability standards, and manufacturing a page through self-editing or paid proxies is against the project's rules, regularly detected, and publicly embarrassing in a way that itself becomes training data.
The legitimate work is narrower and still valuable: make sure pages that already mention you, your category's page, lists you appear on, are accurate, raising corrections through talk pages with sources rather than edits from the marketing department. And treat notability as an output: the independent coverage that would justify a page is the same coverage the mention currency wanted anyway.
The Reddit lane: be a citizen, not a campaign
Reddit's citation weight makes every thread naming your category a small ballot, which is exactly why the clumsy version fails: communities detect promotion with practiced hostility, moderators ban it, and the receipts stay searchable, and now quotable, forever.
The version that compounds: real participation under a disclosed identity, mostly helping with no agenda, mentioning your product where it genuinely answers the question and saying who you are. Monitor threads where your category gets compared, and treat recurring complaints about rivals as product intelligence, not ad space.
The line to never cross is manufactured consensus: sockpuppets, fake reviews, vote rings. That was always bannable; at model scale it is poisoning a well that remembers.
The YouTube lane: publish the workshop
YouTube's citation role runs through transcripts: spoken expertise becomes readable text, and assistants quote it for how-to and review intent. Google's own surfaces lean on it hardest, which the three-surfaces map predicts.
The playbook is publishing's oldest: make the tutorials your niche searches for, demonstrate honestly, and let creators review the product independently, seeding review units without scripting outcomes. Podcast appearances count double here, one conversation becoming transcript text on multiple platforms.
The meta-rule across all three
Each surface's currency is authenticity, enforced by its community and now amplified by the models reading it. Every shortcut, the bought page, the sockpuppet thread, the fake review, converts a compounding asset into a permanent liability the moment it is caught.
Which means the strategy is the same on all three lanes: earn presence at the speed trust actually moves, and let the sampled measurements confirm the associations landing over quarters, not news cycles.
The one-line takeaway: Wikipedia defines you, Reddit vouches for you, YouTube demonstrates you, and the models read all three. Keep the encyclopedia accurate, be a disclosed citizen of the forum, publish the workshop, and never buy what these surfaces only respect earned.