Two GEO Experiments Challenge AI Visibility Advice

Zeeshan Yaseen ran two GEO (generative engine optimization) experiments, and the results didn't match common assumptions. Both experiments tracked 15 commercial-intent keywords across multiple AI platforms.

The first experiment ran for several months and focused on a consultant's brand. Yaseen searched directly on four platforms, ChatGPT, Claude, Gemini, and Perplexity, turning a VPN on and off to check for differences by location.

Citation counts varied widely by platform. ChatGPT led with 148 citations, followed by Claude with 96, Gemini with 87, and Perplexity with 64. Keyword presence peaked at 37.01% on April 29.

Breaking down citation sources, listicles accounted for the largest share at 72.4%, followed by PR at 24.1%, with the remainder split between guest posts, owned sites, and LinkedIn. Notably, a single SEO listicle on Indeed was cited 190 times, more than every other source combined.

The second experiment tracked a SaaS link-building agency starting from a cold start, running 30 days from May 30 to June 28. This time, six platforms were checked manually: ChatGPT, Claude, Gemini, Google AI Mode, Grok, and Perplexity.

Appearance counts by platform were Gemini with 104, Google AI Mode with 95, Claude with 59, and ChatGPT with 32, while Grok and Perplexity trailed at 4 each. The time from publication to first citation ranged from as little as one day to as long as 18 days.

During this period, 18.5% of new users came through referral traffic, and 3.25% came through GA4's AI Assistant channel. The top citation sources were Indie Hackers (146 mentions, up 232% after placement), Bruce Jones SEO (69 mentions, up 165%), and TechBullion (37 mentions, up 147%).

The second experiment changed Yaseen's own thinking. He said "an owned listicle is more of a foundation than a primary growth engine," and the numbers backed that up: third-party listicles accounted for 85.8% of mentions, compared to just 14.0% for owned content.

Comparative content clearly outperformed content that simply advocated for the brand. In one case, adding competitor names to an owned listicle increased mentions by 12.25 times.

Yaseen found that source authority and relevance mattered more than content quality alone, with citations tending to favor government and educational sites, then news sites, then industry sites, then general sites, in that order. About half of cited sources stopped being cited within 30 days, and citation volume didn't correlate with actual referral traffic: Indie Hackers had the most citations but little referral traffic, while TechBullion had fewer citations but more referral sessions, at 64.

In Korea

Neither experiment specifies a brand, market, or language tied to Korea, so whether these findings apply to Korean-language AI search can't be confirmed from this material alone. Anyone attempting GEO in Korea would need to track citations across their own channels directly rather than relying on these results.

Source: Search Engine Land, Zeeshan Yaseen, September 14, 2026. Checked September 16, 2026. Read the original