TL;DR
- Brand sites generate 57% of citations globally across all models, but Gemini relies on them 69% of the time while ChatGPT uses them least at 47%.
- Industry norms vary dramatically: pharma sees 33% brand citations versus 72% in marketing software; cybersecurity is highest at 74% and government/non-profit lowest at 15.9%.
- Know which citation buckets drive your sector, then publish or partner to shift the mix: social media drives 16.5% of citations in fashion but only 5% in pharma.
Citation-mix is different for every industry
Citation composition reflects how each industry produces content. Pharma and biotech companies, whose own content passes through regulatory review and multi-party approval, see a median 59% of citations go to earned media. SaaS companies, which publish constantly, see only 11.4% earned. The mix an engine uses is shaped by what the web makes available.
Where the information comes from depends on which model answers
Each LLM has its own information universe, shaped by search partners and data licensing. ChatGPT leans most on earned media (30% of citations); Google AI Overviews least (17%). Google's surfaces cite social sources heavily, AI Overviews at 15.3% and AI Mode at 14.4%, while Microsoft Copilot sits near 3.4%.
What does this mean for marketers?
Industry benchmarks are baselines, not ceilings. Within a single industry the gap between the 75th and 25th percentile company can exceed 43 percentage points of earned-media citation share. A pharma brand with 30% earned citations is underperforming its 59% median; a fintech brand at 30% is beating its 15% median by double. Competitive position only makes sense inside the category.
| Industry | P25 | Median | P75 |
|---|---|---|---|
| Pharma / biotech | ~38% | 59% | ~81% |
| Fintech | ~7% | 15% | ~31% |
| SaaS | ~5% | 11.4% | ~24% |
Understand your industry's citation mix, identify which buckets drive it, then publish strategically or partner with the publishers that engines already trust.
Methodology
We analyzed 11.84 billion citations across 8 models (ChatGPT, Claude, Google AI Mode, Google AI Overviews, Google Gemini, Grok, Microsoft Copilot, Perplexity) from April to July 2026, covering 8,061 categories (7,542 with at least 100 citations) in 29 industries. Domain classification mapped 3.02 million domains (98.3% coverage) into three buckets: brand, earned and social. Categories were weighted equally regardless of citation volume; regional and demographic variation is out of scope.