Does Luxury Presence do AEO and help you get found in AI search?

The cohort contained 102 selected Luxury Presence websites. Each site received five prompts per engine across ChatGPT, Gemini, Perplexity, and Grok: one branded question and four non-branded questions. That produces 20 answers per site, with 2,040 total answers in the historical panel.

The branded prompt asked about the agent by name. Non-branded prompts asked for an agent serving a market or need. The study used a stricter site/domain recommendation rule than a simple business-name match: a URL appearing only as a scraped citation did not qualify as a recommendation. Strict domain scoring can miss an agent recommended under a brokerage or another domain.

The historical branded measure recorded 125 of 408 answers (30.6%). The non-branded recommendation measure recorded 134 of 1,632 (8.2%). The same reader should not interpret these as equivalent measures of new-customer discovery.

A consumer who already knows an agent can ask for that agent directly. A consumer building a shortlist asks whom to consider. The gap between those tests is useful when deciding what to measure, but this study does not isolate why the gap exists or whether a different platform would change it.

The original report describes ChatGPT and Gemini as trained-memory modes, Perplexity as live web, and Grok as mixed. Those are historical test labels, not descriptions of every current product mode. Engine results below retain the original answer denominators.

Three of 102 sites had at least one qualifying ChatGPT recommendation, and three had at least one Gemini recommendation. The corresponding answer-level counts were 7 of 408 and 8 of 408. The overlap of the two sets of three sites is not established, so they cannot be combined into six unique sites.

Test discovery by need as well as recognition by name. Document whether a recommendation points to your website, a brokerage profile, a directory, or another source. If an engine names you but does not cite your domain, that is a different finding from not naming you at all. Preserve the distinction when comparing vendors or setting a baseline.

Use the date as part of every claim. A June or July result cannot describe a vendor's November performance, current product, or customer services. A later retest needs the same selection rules, prompts, scoring contract, and eligible-answer handling before it can support a change-over-time claim.

A useful website should make its content accessible, its entities clear, and its evidence easy to inspect. Google Search Central says established SEO practices remain relevant to AI Overviews and AI Mode, including crawl access, internal links, textual content, and structured data that matches the visible page. That guidance supports a technical review; it does not prove that a missing field caused the recommendation rate in this study.

AEO, GEO, and LLM visibility work should therefore start with a measurable consumer question and traceable source material. Publishing more pages without knowing which questions are unanswered or which claims need corroboration can increase volume without establishing useful evidence.