Two GEO experiments challenge standard AI visibility advice

Two new generative-engine optimization experiments, reported by Search Engine Land, directly challenge the conventional advice agencies now sell as AI visibility strategy. That is the hardest fact in this story: practitioners ran structured tests, and the results did not match the standard playbook.

The detail matters here. Search Engine Land's report, titled "Two GEO experiments challenge conventional AI visibility advice," describes a pair of tests aimed at generative engine optimization — the practice of shaping content so large language model search surfaces cite it. We do not yet have the full methodology in front of us. That gap deserves scrutiny before anyone rewrites a strategy document.

What do the experiments actually test?

According to the report, the experiments examine widely repeated GEO claims — the tactics consultants prescribe for appearing in AI-generated answers. These typically include structural changes to content, formatting adjustments, and citation-friendly phrasing. The experiments asked whether those tactics hold up under controlled comparison.

The reported answer: partially, at best. The tests found the conventional guidance did not deliver as promised across the board.

A caveat belongs beside every promise in this field. The Search Engine Land piece summarizes the experiments but the underlying test design, sample sizes, and query sets are the decisive evidence, and readers should weigh those before acting. A single experiment, or even two, cannot settle how generative engines select sources at scale. Takeaway: tested claims beat untested claims, but two tests do not make a science.

Why does this matter now?

Generative engine optimization has grown quickly from conference talk line item to billed service. Vendors promise AI visibility the way they once promised featured snippets. Experiments that contradict the standard advice arrive at an awkward moment for that market.

Sites that invested in GEO tactics on faith now face a simple question. Did those changes cause AI citations, or did correlation masquerade as causation? These experiments suggest some of the assumed cause-and-effect does not survive testing. The report does not claim GEO is useless; it claims specific, popular recommendations lack support.

That distinction matters for publishers, e-commerce sites, and local businesses alike. Verticals that rushed to restructure content for AI answers may want to hold further spend until replication confirms which tactics carry real effect. Takeaway: the burden of proof now sits with the playbook sellers, not the skeptics.

What should readers watch next?

Separate the confirmed from the speculative here. Confirmed: two experiments exist, they target conventional GEO advice, and Search Engine Land reports their findings challenge it. Speculative: whether these results generalize across engines, languages, and verticals. Nobody has shown that yet.

Watch for three things in coming weeks. First, publication of full methodologies, including query samples and control conditions. Second, independent replication by parties without a GEO service to sell. Third, any response from the vendors whose guidance these tests undercut.

The open question is durability. If follow-up experiments confirm these findings, the AI visibility consulting market will need to revise its claims quickly. Monitor Search Engine Land and the original researchers for methodology releases before treating any single GEO tactic as settled fact.