Zeeshan Yaseen ran two GEO experiments, and third-party listicles produced 85.8 percent of AI citations

Zeeshan Yaseen, an SEO consultant and founder of ZeeKnows, published two hand-tracked GEO experiments on Search Engine Land on 14 September. The first ran over several months across ChatGPT, Claude, Gemini and Perplexity and logged 395 citation events on 15 commercial keywords. The second was a 30-day cold-start test across six platforms, adding Google AI Mode and Grok, that logged 298 citation events on 15 more keywords.

What the split looked like

In the first experiment, listicles produced 72.4 percent of citations and PR produced 24.1 percent, with guest posts, the owned site and LinkedIn splitting the rest. Yaseen writes that the second experiment sharpened the picture: of 437 source mentions on the tracked keywords, third-party listicles produced 85.8 percent, the brand's own listicle produced 14.0 percent, and PR produced 0.2 percent.

The lesson Yaseen draws is that an owned listicle is a foundation. It is the slowest source to be cited, and once it is cited it carries a small share of the answer. Earned placements on third-party listicles do most of the work.

The competitor-comparison test

On 23 June, mid-experiment, Yaseen updated the brand's own listicle so that it compared the brand to named competitors instead of describing the brand alone. Mentions of that page rose from 4 to 49, which Yaseen reports as a 12.25-fold increase. The signal is that entity associations matter: pages that name the brands a reader is already comparing get pulled into the model's answer more often than pages that stand alone.

Sources decay in weeks, not months

Roughly half of the sources the models cited stopped being cited within 30 days, without any new placement or refresh on the site's part. Yaseen writes that this makes a one-time PR push a poor fit for GEO work: the pipeline needs continuous placements to keep the same set of sources fresh in the answer.

The traffic side is smaller than the citation side. Yaseen reports that 18.5 percent of new users on the second experiment's site came in through referral traffic and 3.25 percent through GA4's AI Assistant channel, a little over one fifth of all new users combined. Citation volume did not track referral volume: the most-cited sources were not the strongest referral sources.

Two caveats, from the article and worth stating plainly. Both experiments were run by the author's own consulting practice on brands they were working with. Data was logged manually rather than through a citation-tracking platform, so the sample sizes are small and one person's coding decisions sit behind every number.

For a team trying to be cited by ChatGPT, Claude or Perplexity, the advice that comes out of the numbers is: check which third-party listicles the models already cite for the target queries, budget for ongoing placement on those listicles rather than for one-off PR, add a comparison page against the brands the audience is already weighing, and measure referral conversions rather than citation counts.

Source

Two GEO experiments challenge conventional AI visibility advice, by Zeeshan Yaseen, Search Engine Land, 14 September 2026.