We asked ChatGPT and Perplexity to recommend a fee-only financial advisor in 100 cities across the US and Canada, three ways each, and read all 600 answers. Every answer named specific firms, 7.9 on average. But the two assistants agreed on fewer than 1 in 5 of the firms they named for the same city, and rewording the question changed most of the list. The answers were built from fee-only directories, consumer finance sites, review sites and firms’ own websites, so what a firm publishes about itself, and where, decides whether it can be named at all.
How we ran it
On October 31, 2026 we asked each assistant three questions in each of 85 US and 15 Canadian cities:
- “I’m looking for a fee-only financial advisor in [city]. Who would you recommend?”
- “Who are the best fee-only financial planners in [city] for retirement planning?”
- “I want a flat-fee financial advisor in [city], not one who charges a percentage of assets. Which firms should I look at?”
That’s 300 answers from ChatGPT, with web search, and 300 from Perplexity. We recorded every firm, advisor and directory each answer named, and every site it cited. The full method and its limits are at the end.
Every answer names firms, but rarely the same ones
Neither assistant declined to name names. Every answer listed specific firms or advisors, from 2 to about 20 of them. Across the 600 answers they named 2,622 different firms, and most came up once or twice.
Source: Wellspent AI recommendations study, October 31, 2026 (600 answers)
For a person choosing an advisor, the list depends on which assistant they open. In the average city, only 18% of the firms named by either assistant were named by both. In three cities the two lists had nothing in common.
The wording changes the answer
The list moves even when the assistant and the city stay the same. Asking for a “fee-only financial advisor” and asking for “the best fee-only financial planners for retirement planning” in the same city, from the same assistant, produced lists that overlapped by 28% on average.
That’s the real behavior to plan for. Clients don’t type three-word searches; they describe their situation, and every situation is a slightly different question. A firm named for one phrasing is often absent from the next. The firms that show up across phrasings are the ones whose pages answer the specific question: the client type, the fee model, the situation.
Where the answers come from
Both assistants cited their sources, and they read very different parts of the web. ChatGPT cited about 15 sites per answer and leaned on fee-only directories and the SEC. Perplexity cited about 9 and leaned on consumer finance sites, rankings and review sites.
A firm that’s well represented in one assistant’s sources can be invisible to the other. Being findable now means being present, with consistent facts, across several kinds of source rather than one.
Firms’ own websites are part of the answer
77% of answers cited at least one fee-only firm’s own website: 87% of ChatGPT’s and 66% of Perplexity’s. The assistants read firm sites directly and quote what they find there, which makes a firm’s site a source in its own right, not just the place a client lands afterward.
That’s where our website study matters. Most fee-only firm websites say the same three things (retirement, fee-only, fiduciary), about half publish a price we could verify, and about 4 in 10 name no specialty beyond the general ones. An assistant answering a specific question has little on those sites to match.
The lists get read. The firms get the call.
Lead-generation sites are among the most-read sources, but readers don’t stop there. SmartAsset was cited in 48% of all answers, almost always its city pages of top-rated advisors, yet the assistants recommended it by name in none. They read the lists for firm names and sent readers straight to the firms.
Assistants also look for verified firm facts. In ChatGPT’s answers, Wellspent’s public graph at find.wellspent.co was cited in 37%, more often than SmartAsset (33%) or Yelp (32%), and in Canada it was cited as often as MoneySense (39% of answers). It was read for the facts it holds about each firm: fees, specialties, the clients served, and SEC registration. Perplexity, which leans on consumer and review sites, rarely cited it.
Most of the firms the assistants named already have a record there: 73% of the firms ChatGPT named in US answers, and 57% of Perplexity’s.
Assistants tell people to check
40% of answers told the reader to look the advisor up on the SEC’s Investment Adviser Public Disclosure site before hiring, and 25% mentioned NAPFA. Assistants increasingly send people to verify, which rewards firms whose registration, website and listings all say the same thing.
Canada looks different
Canadian answers named fewer firms, 6.4 on average against 8.2 in the US, and cited firms’ own websites far less often: 20% of Canadian answers against 87% in the US. They leaned on Canadian sources instead: MoneySense (cited in 39% of Canadian answers), fee-only directories such as feeonlyfinancial.ca and maketheswitch.ca, and the public graph (39%). For Canadian fee-only planners, the directories and lists the assistants read carry more of the weight.
What this means for a fee-only firm
-
Answer specific questions on your own site. Assistants match situations, not slogans. Name the clients you serve, the situations you handle and how you charge, in plain text.
-
Publish your price. Assistants quote prices when they can find them. A firm without one can’t be the answer to “a flat-fee advisor who charges about $X.”
-
Be consistent everywhere assistants read. The same fees, specialties and service area on your site, your directory listings and your SEC filing.
-
Check more than one assistant. Being named by one tells you little about the other. Ask both, with a few different phrasings of your ideal client’s question.
-
Be a source, not a listing. Assistants read lists and send people past them to the firm. What gets a firm named is having accurate, specific facts where assistants look, and the detail behind those facts carries the rest of the conversation. When an assistant cites a page, it pulls out what that page says: the fee model, who the firm works with, the situations it handles, its registration. The client’s next question then gets answered from those facts. “Do they work with physicians?” “Would they handle selling my practice?” “What would a plan cost?” A record that only says “fee-only, retirement planning” gives the assistant nothing to work with past the first question.
That’s why Wellspent works with each member to fill in their record beyond what public filings show: the clients they serve best, the situations they know well, how they charge and why, in their own words. That nuance is what an assistant takes from the record when it cites it, and what it relies on as the conversation narrows from “a fee-only advisor in Denver” to “the right one for me.”
For the background on how assistants choose, see How AI assistants choose which financial advisors to recommend, and for the steps, How to get your firm recommended by AI.
Methodology
Questions. Three questions, quoted above, in 100 cities: the 85 largest and best-known US metros we cover and 15 Canadian cities. Each was asked once of each assistant on October 31, 2026, for 600 answers.
Assistants. ChatGPT with web search, and Perplexity. We recorded each full answer and the sources it cited.
Names. We used Claude to list every firm, individual advisor, directory and matching service each answer named, and whether the answer called it fee-only. Counts of distinct firms use normalized names, so minor spelling differences count as one firm.
Overlap. For each city we compared the set of firms each assistant named across its three answers, and report shared firms as a share of all firms named by either (Jaccard overlap). The wording comparison uses the fee-only and retirement questions from the same assistant and city.
Sources. A site counts once per answer however many of its pages were cited. Image and map hosts are excluded.
Records. To count named firms with a record in the public graph, we matched firm names (not individual advisors) to the graph’s firms by exact normalized name. Strict matching misses some name variants, so the true shares are likely higher. Canadian matches were too sparse to report.
Limits. Answers vary from one run to the next, so this is a snapshot of one day. We didn’t check whether each named firm is in fact fee-only; many answers named individual advisors rather than firms, which makes matching to registrations unreliable. City choice leans toward larger metros. Wellspent runs find.wellspent.co, one of the sites cited.