What the citation data shows
AI systems cite social content at a fraction of Google's rate. Researchers at the University of Toronto classified every source cited by four AI systems and Google across 300 queries as brand-owned, earned (independent editorial) or social (user-generated and community platforms), in a study presented at the EDBT/ICDT 2026 workshops.
| System | Earned | Social | Brand |
|---|---|---|---|
| Claude 4.5 Sonnet | 65% | 1% | 34% |
| GPT-4o | 57% | 8% | 35% |
| Perplexity Sonar Pro | 50% | 11% | 39% |
| Gemini 2.5 Flash | 46% | 8% | 46% |
| Google Search | 41% | 34% | 26% |
The gap widens where it matters most. For consideration queries, the vendor-evaluation stage, the AI systems converged on earned media at 59–86% of citations with social close to absent, while Google leaned on social at 41%. Social content is strongest in Google exactly at the stage where AI uses it least.
Why assistants avoid social sources
Assistants avoid social content because it is hard to synthesise reliably. An answer has to be condensed into a few attributable sentences. An editorial article has a named author, a structure and an argument; a Reddit thread has replies, disagreements and community context that only make sense inside the thread. Choosing the structured article over the thread is a quality decision, not an oversight.
Three other mechanisms reinforce it. Credibility signals: institutional publications carry accountability markers that anonymous posts lack, and systems trained for reliability weight those markers. Legibility: Ubiratan de Oliveira's analysis of the shift from SEO to GEO describes generative legibility, content that both people and inference processes can interpret, and social content scores low on the machine side. And architecture: Sinan Aral and colleagues at MIT found in their study of AI search that AI answers show far lower response variety than traditional search results across 2.8 million results in 243 countries. A one-voice answer is built from the most consistent sources available, and social feeds are the least consistent.
None of this is a temporary state of the technology. The pressure on AI systems is toward more reliability, which means more editorial and structured sources, not fewer.
What this means if your visibility is built on social
A social-first visibility programme can look healthy in every dashboard and still be invisible where buyers now ask. Follower counts, engagement and reach keep rising while inclusion in AI answers stays at zero, and no social analytics tool shows the gap.
The commercial exposure is specific. Aral and colleagues report that 67% of US search queries were answered by AI in 2025, with business, finance and employment queries up 69% in AI coverage year over year. Those are the queries where vendor evaluation happens, and the answers are built from editorial and brand sources. The AI search behaviour piece covers how the buying journey moved.
Social media still does real work: community, relationships, referral traffic, direct engagement. Its role in AI search is indirect and small: sharing editorial coverage may lift its Google performance, which marginally helps retrieval; Gemini's Google grounding gives social a slightly larger share (8%) than Claude's 1%. Neither replaces the sources AI actually cites. Budget social for what it does, and build the separate programme for what it cannot do.
Where to move the effort
The Toronto source shares give an investment order for AI visibility:
- Earned editorial coverage, 46–65% of AI citations. Mentions in the specific outlets Perplexity cites for your category's consideration queries, with descriptions that say what you do and for whom. The digital PR piece covers the method.
- Brand-owned structured pages, 26–46% overall and 52–68% for purchase-intent queries. Service pages with scope, timeline and FAQPage schema.
- Entity foundation. Organization schema and consistent naming across the web, which Kargaev's ranking-factor analysis identifies as the strongest single GEO signal (NIS 0.918). Without it, coverage and pages are cited without your name.
- Social, 1–11%. Funded for community value, not for AI visibility.
The skills partly transfer. Knowing what buyers ask, how to phrase a position and what a good story is all carry over. The implementation does not: schema, entity alignment and FAQ architecture are a different craft, laid out in the GEO checklist.
Frequently asked questions
Why does Google cite social content so much more than AI does?
Google is built to surface a diversity of sources, including community discussion, and has features that deliberately index and rank it. AI assistants are built to produce one reliable, synthesised answer, and community content is the hardest source type to summarise accurately.
Does LinkedIn content count as social?
Yes, in the Toronto classification: individual posts and organic content are user-generated without independent editorial oversight. A structured company page sits closer to brand-owned content. The same effort pitched to an industry publication typically earns more AI visibility than publishing it directly on LinkedIn.
Should I stop investing in social media?
No. Keep it for community, relationships and referral traffic, and stop expecting it to produce AI citations. The AI visibility programme is separate: editorial coverage, structured pages, entity data.
Is Reddit worth it, given Google's push on discussion content?
For Google organic and for category-level context in AI Overviews, Reddit can help indirectly. For being named as a recommended vendor in an AI answer, it produces very little; brands are almost never cited specifically from threads.
Do review platforms like G2 or Trustpilot count as social?
They sit between social and earned: structured, with review processes and institutional signals that feeds lack. Accurate, specific profiles there are a low-risk supplementary investment, less reliable than editorial coverage but more useful than social posts.
Will AI systems start citing social more over time?
Unlikely. They are evaluated on accuracy and trustworthiness, and unverified community content works against both. The direction of travel is toward more editorial and structured sources.
See what your social presence is producing in AI answers
The free analysis checks whether ChatGPT, Perplexity and Gemini name you for your category today, and which sources they cite instead.
