At brightonSEO 2026, Neil Patel shared findings from 9 million AI answers tracked across 9 platforms and more than 400 enterprise brands, spanning ecommerce, finance and B2B software. The dataset tests nine assumptions that keep showing up in GEO pitches, LinkedIn posts and Slack threads.
The throughline: many teams are spending on the wrong signals. Sentiment, format and platform-specific tricks look good in a pitch but do not survive contact with the numbers.
Nine assumptions that crumble under real prompts
Here are the findings that matter most for growth teams and agency operators.
- Sentiment has a negative or flat correlation with citations. Discussion volume feeds retrieval; being quietly well liked does not.
- 64% of AI citations come from ordinary pages, not listicles, how-tos or comparisons.
- On commercial-intent prompts, 82% of citations go to third parties and only 3% to owned pages.
- YouTube delivers the strongest third-party visibility lift at 2.8x, ahead of LinkedIn at 2.7x and Reddit at 2.4x.
- Owned citations are rare but produce a 5x visibility jump when they happen.
Two more data points deserve attention. Up to 58% of citations never reappear, so a single citation spike is a weak target. And up to 75% of AI answers that cite a brand’s page never actually name the brand.
What actually drives AI citations
The stable high performers in the study tend to share two characteristics: comprehensive topic coverage and meaningful third-party references. That explains why brand-level visibility swings are much wider than platform-level swings. When AI visibility dips, the first audit should be your own content and authority, not the model update calendar.
Owned content remains valuable, but not for the reason most assume. Owned citations are rare on money prompts, yet when a brand’s own page gets cited, its AI visibility rate jumps 5x. Those owned citations also last 3 to 9 times longer than third-party ones.
Three moves for your GEO plan
First, aim digital PR at bottom-of-funnel topics such as pricing comparisons, category roundups and best-for placements. Those are the commercial-intent queries where third parties dominate.
Second, treat YouTube as a citation surface. A video transcript is a large spoken-language document that models retrieve well. Add full transcripts, descriptive titles and consistent brand mentions inside the video, not just in the description.
Third, change the metric. Citation count alone is not enough when many citations never name the brand. Track AI visibility, share of voice, branded search and direct traffic.
The citeability framework
Think of AI visibility as three layers. Topical depth makes your brand eligible for an answer. Third-party coverage signals that the market treats you as a reference. Owned pages then turn that eligibility into durable visibility when the model points back to your domain.
Chasing only one layer produces spikes. Building all three is what separates the 4x steadier top performers from everyone else.
Source: Neil Patel
