Luxury jewelry brand, U.S. market · three-month GEO program · Client details anonymized. Program metrics are actual.
Situation
The brand sells fine jewelry to American customers who research before they buy. A growing share of that research now happens in ChatGPT, Claude, Gemini, Perplexity and Google's AI Overviews, in private conversations no analytics dashboard records.
Two years earlier, the brand had gone through a rough patch on after-sales service: slow repairs, unanswered emails, a run of complaints on forums and in archived articles. Operationally, the problem was fixed. New team, new process, new numbers.
The AI engines had not caught up. All five major models still drew their answers from the old complaints. Two competitors with weaker service records were positioned as the category's service benchmarks, on thinner evidence. When a prospect asked an assistant "is this brand reliable after the purchase", the answer came from 2024.
Leadership asked a simple question: what do the AI engines actually say about us, where do they get it, and can we change it?
Method
Measurement first. We ran a few hundred brand-relevant prompts across the five models, the questions a real buyer asks: reliability, heritage, resale value, after-sales, comparison with named competitors. For each answer we recorded the position taken, the sentiment, and the sources cited. Each model told a different story, so each was scored separately.
Source mapping. The central finding of any GEO audit: the sources that shape AI answers are rarely the ones a brand invests in. Here, the brand's own website was cited in 13% of answers. Forums, Reddit threads, archived articles and third-party reviews carried the rest. Several of the most cited threads were years old and closed to new replies.
Competitive gap. The same prompts, run for the two direct competitors, showed exactly where the brand was absent, misframed or outranked, model by model.
Remediation. Three months of GEO work on the sources the engines actually read: authoritative pieces placed in GQ, luxury media and trade publications; heritage and craftsmanship content built to be cited; technical groundwork on the site so the engines could read it; and fresh, factual presence where the old threads lived. No argument inside the old complaints. Better sources, placed where the engines look, displace them.
Monitoring. AI visibility tracking now runs alongside the brand's traditional media monitoring, with monthly prompt simulation to catch new negatives early.
Results
After three months:
- Brand website citation rate: 13% to 30%. The brand's own pages became a source the engines trust.
- Heritage citation rate: 15% to 50%. The story the brand wanted told, told by the models.
- After-sales negatives: 11 to 0. Resale negatives: 6 to neutral.
The two competitors lost their "service benchmark" framing in the same period, not because anything was said against them, but because the evidence base changed.
What this means for any luxury brand
Three lessons a board can act on without technical expertise.
Your reputation now lives inside the models, and you cannot see it from your dashboard. Social listening and press monitoring watch public channels. AI answers are private, one buyer at a time. If nobody is measuring per-model sentiment, nobody knows.
AI engines compile history. A problem fixed operationally stays alive in the answers until the sources change. Reputation work that stops at press coverage leaves the most durable layer unmanaged.
Distributed content beats owned content. Reddit, Wikipedia, LinkedIn and third-party editorial drive AI citations far more than brand websites. Luxury brands still optimizing their own domain for Google are optimizing the wrong asset for the wrong channel. Gartner projects organic search traffic down 50% by 2028.
Three questions for any CEO this quarter: What are we doing to track AI citations, not just organic traffic? Which sources do the models cite about us, and do we have a presence there? Which of our pages sit at positions 10 to 20 on Google and have not been refreshed this year? That last one is where citation investment compounds fastest.
Sources: