Nearly half of ChatGPT's advisor citations were advisors' own websites

Bonfire Studio, a website design and digital strategy firm, ran advisor-seeking queries through ChatGPT across five metro areas and catalogued the 1,048 sources the model cited back; 45% of those sources were websites owned by individual advisors or advisory teams ^[1][2][3]. That puts the asset most firms build once and then forget ahead of every other input into what a language model says when a prospect asks it who to call.

The other categories reported came in far lower. Government and regulatory agencies together with professional associations were returned in nearly 11% of searches, and sites owned by banks, wirehouses and insurance companies accounted for 8.2% ^[4][5]. Taken as shares of the same pool, advisor-owned sites showed up roughly five and a half times as often as the large-institution sites ^[6]. The denominators in the published figures are not consistent, though: the 45% and the 8.2% are described as shares of sources, while the 11% is described as a share of searches ^[1][4][5]. On the source count, 45% of 1,048 is about 470 citations ^[7].

The most cited single domain was not an advisor's. It belonged to the National Association of Personal Financial Advisors, a trade group ^[8]. So the aggregate lever is your own site, while the biggest individual lever is someone else's directory.

Bonfire's president and founder, Sara MacQueen, said the first step to getting cited is having blogs or other writing online that shows expertise on topics such as tax planning or working with high-net-worth clients, and that posting once is not enough ^[9]. She said large language models are known to favor information published in the previous six months ^[10], and that the common client pattern is a site built at launch and then treated as finished while advisors join and leave, locations open, and the design ages ^[11]. Keep the conflict in view: Bonfire sells website design and digital strategy, so a finding that websites are the highest-value channel is convenient for Bonfire ^[2].

The less flattering result is reproducibility. When Bonfire re-entered its Charlotte queries a second time, the websites ChatGPT cited changed nearly 45% of the time ^[12], which means only about 55% of cited sites held their place on an identical prompt ^[13]. MacQueen said AI systems are known to work from probabilities ^[14]. For operators, that is the number that should govern spending: any vendor report showing you cited once in an AI answer is a coin flip, not a ranking.

Fran Merlie, senior vice president of content and AI strategy at the marketing firm Gregory, agreed that firms racing for AI visibility neglect their own sites, and said models tend to consult a wide variety of sources when gauging authority ^[15][16]. His practical complaint was blunter: it is often not clear from a company's own site what the company does ^[17]. Older tactics, such as packing pages with likely search keywords, are losing their power to secure top placement ^[18].

What to watch: the study tested ChatGPT only ^[19], while the same visibility questions apply to Anthropic's Claude, Google's Gemini and other models ^[20], so the 45% share is one engine's citation mix rather than a general rule. Watch also whether the six-month freshness preference MacQueen describes shows up in your own logs, and treat single-snapshot citation audits as noise until someone publishes the variance.