A search consultant who tracked AI citations by hand across two separate tests found that where a brand gets mentioned matters more than what it publishes on its own site. Zeeshan Yaseen described the work in a Search Engine Land contributor piece published September 14, running the tests back to back on two different brands and logging 775 citation events across ChatGPT, Claude, Gemini, Perplexity, Google AI Mode and Grok. The samples are small and self-run, not a controlled study, but the pattern that emerged is worth tracking.
The first test followed 15 commercial-intent keywords across four platforms for an existing consulting client over several months. The brand ended up appearing for 10 to 12 of the 15 terms, with keyword presence peaking at 37.01% on April 29. ChatGPT produced the most citations (148), ahead of Claude (96), Gemini (87) and Perplexity (64). A single listicle hosted on Indeed SEO drove 190 mentions, far more than any other individual source.
Yaseen observed that listicles, PR and guest posts reinforced one another rather than acting as separate channels: a press release amplified a listicle’s claim, and guest posts referenced both. The test also turned up a peer-set effect. One listicle placed the brand alongside Lily Ray and Aleyda Solis, and citations dropped within days after those names were pulled, which suggests the models weigh the company a brand keeps, not only the individual mention.
Roughly half of the cited sources in that first test stopped appearing within 30 days, and one placement dropped from 29 mentions to 11 within a week. Coming out of that first test, Yaseen recommended publishing listicles on the brand’s own site, because many top-ranking brands appeared to benefit from their own content.
The second test started from zero. A SaaS link-building agency had no measurable presence on any of the six tracked platforms between April 30 and May 29, the test’s baseline period. Over the following 30 days, the same 15 commercial-intent keywords produced 298 appearances, led by Gemini (104) and Google AI Mode (95), with Claude, ChatGPT, Grok and Perplexity well behind. Referral traffic plus GA4’s AI Assistant channel made up slightly more than 20% of new users during that window.
That result forced Yaseen to revise the conclusion drawn from the first test. The brand’s own listicle needed 18 days to earn its first citation, slower than every third-party placement, and it contributed only 14% of the 437 tracked mentions against 85.8% from earned third-party listicles. Rewriting the owned page to cover competing agencies rather than promote the brand alone lifted its citation count from four to 49, but Yaseen now describes an owned listicle as a foundation rather than a primary growth engine.
Both samples are tied to two brands the author worked on directly, so the citation counts describe what happened in these two windows rather than how AI platforms rank sources in general. Teams building a GEO program can still test the underlying method on their own keywords: run target terms through the relevant AI platforms before choosing outreach targets, letting the sources already being cited shape the outreach list, which Yaseen notes may look very different from a traditional prospecting spreadsheet.
Search Engine Land published Zeeshan Yaseen’s account of both experiments on September 14, 2026.
