I Scored 368 AI Answers About Nashville Advisors. My Firm Was in 57% of Them.
Sixteen questions, eight times each, to three AI assistants. 384 requests, 368 answers that came back, every one of them read and scored.
My firm's website was cited as a source in 211 of them, which is 57%. My firm was named in 230, which is 62%. September 2026, one firm, one metro, one month.
The 57% is the headline and it is the least useful number in the report. What mattered was the shape of the other 43%, and one uncomfortable thing about the 57% itself that I did not expect to find.
Cited as a source
57%211 of 368
The answer used my website as one of its sources.
Named in the answer
62%230 of 368
The answer mentioned my firm by name.
Melby Wealth Management, September 2026, first measured month. Cited and named are separate measurements and are never blended.
Why asking ChatGPT once tells you nothing
Type "who are the best fee-only financial advisors in Nashville" into ChatGPT right now. You will get three or four firms. Type it again tomorrow and the list shifts. Ask Claude and it shifts further. Ask from a laptop in another city and it changes again.
That instability is why a single check is worthless and why a rate means something. One answer is an anecdote. Several hundred answers to a fixed set of questions is a measurement.
16
questions a prospect in my metro would ask
8
runs of each question, per assistant
3
assistants: Claude, ChatGPT, Perplexity
368
answers scored, of 384 requested
Sent through each provider's API with web search enabled and Nashville passed as the asker's location. Four brand-check questions are scored separately and excluded from these counts.
So I built a question set a prospect in my metro would plausibly type, sent each question to Claude, ChatGPT, and Perplexity through their APIs eight times each with web search enabled and Nashville passed as the asker's location, and scored every answer that came back.
The set covers four kinds of discovery. Broad: an advisor in Nashville, with no other qualifier. Niche: an advisor who works with a specific kind of client. Service: who does a particular kind of planning here. Comparison: at this asset level, a large firm or an independent RIA. Four more questions ask about my firm by name and are scored on a separate line and kept out of the headline, because a model knowing who you are is a different thing from a model recommending you.
Sixteen discovery questions, 23 answers each. 368 scored answers.
The headline, and the number under it that matters more
Cited in 57% of 368. Named in 62% of 368.
Those are separate measurements and blending them would be dishonest. Named means the answer mentioned Melby Wealth Management. Cited means the answer used my website as one of its sources. The second is harder to earn, so it leads.
Now the figure I actually watch:
Cited rate by question type, Melby Wealth Management, September 2026. Brand checks are muted because they are excluded from the headline rate: a model knowing who you are is a different thing from a model recommending you.
Broad discovery: cited in 54% of 92 responses. Named in 65%, which is 60 of 92.
Broad discovery is the plain question with no qualifier. It is the hardest to win, the most people ask it, and it is the number to judge a firm by.
Set that against niche discovery, where I was cited in 98% of 92 and named in 98%. A firm that averaged those two and published the midpoint would be using its strongest result to cover its weakest. My niche pages are doing their job to a degree I did not expect. My general visibility in this city is a coin flip.
The rest: service discovery, 45% of 92. Comparison questions, 33% of 92. Brand checks, 100% cited and 100% named of 92, which reads as recognition rather than discovery and stays out of the headline.
Those five rows sort themselves the same way my website does. The question types where I score well are the ones I have built pages to answer. The ones where I score badly are the ones where a model looking for a page of mine to quote would not find one.
The three assistants disagree with each other
Same month, same questions, same city:
| Assistant | Cited | Named | Model measured |
|---|---|---|---|
| Perplexity | 73% of 128 | 80% | perplexity/sonar |
| ChatGPT | 60% of 112 | 69% | gpt-5.6-luna |
| Claude | 40% of 128 | 40% | claude-sonnet-5 |
Perplexity cited my site nearly twice as often as Claude did, 93 responses against 51. That is a structural difference in how each system decides what to trust, and it is invisible to anyone who checks one assistant and treats the result as a reading of AI generally.
Every row carries the model version measured. When a provider swaps the model underneath you, the comparison with last month stops meaning anything, and the honest move is to say so rather than draw an arrow.
My own site out-cited every directory
Pages cited across the same 368 answers, counted one URL at a time. Directory and competitor names are withheld here; they appear in the report itself. Bars are scaled against my homepage at 196.
Here is the number that changed how I think about this.
Counting one page at a time across those 368 answers, my homepage was cited 196 times. The most-cited directory page in the entire set managed 156.
Directories matter and I am not going to pretend otherwise. I am listed on two of them, a fee-only network page cited 111 times and my own association profile cited 69 times. That is real and I maintain both.
But a citation to my page and a citation to a directory page are not the same event.
A model cites my page
One firm on the page. There is nothing else to name.
The answer says my firm's name.
A model cites a directory roundup
Fifteen or twenty firms on the page. I am one row.
The answer says whichever rows the model picks.
When a model cites my page, it has cited a page about one firm. Mine. The answer names me because there is nothing else on the page to name.
When a model cites a directory roundup, it has cited a page with fifteen or twenty advisory firms on it. I am one row. Which firms come out of that list and into the answer is decided by the model, from a page I do not control, against every other advisor listed there. Being on the page is necessary. It is not sufficient, and it never becomes sufficient, because the competition on that page never goes away.
That is the part I keep not seeing in what gets written about showing up in AI answers. Getting listed somewhere is an afternoon of work with a clear finish line, so it is what gets recommended. Owning the page a model quotes from is slower, and it is the only version where the answer has to say your firm's name.
The pages AI trusts instead of yours, and whether you are on them
The report lists every page the assistants cited, how many of the 368 answers cited it, and whether my firm appears on it.
My homepage: 196 of 368.
A Nashville advisor directory page: 156 of 368. Not listed.
A competitor's own website: 130 of 368.
A competitor's association profile: 120 of 368.
A fee-only network Nashville page: 111 of 368. Listed.
A second competitor site: 104 of 368.
A third competitor site: 92 of 368.
A second advisor directory: 76 of 368. Not listed.
My own association profile: 69 of 368. Listed.
Then I looked at what it would take to be on those two.
The first sits alongside a paid advisor lead product, which is a commercial relationship I decided against for my own practice some time ago. The second picks a fixed number of firms in each city by its own editorial criteria, so there is nothing for me to submit.
Open
A listing you can join, usually tied to a membership or a profile you fill out yourself.
Do these. An afternoon of work, and they stay yours as long as the membership does.
Editorially closed
The page picks a fixed number of firms by its own criteria. There is nothing to submit.
Note it and move on. The citation count is real intelligence. The task is not available.
Declined by choice
A page you could be on, behind a commercial relationship you have decided against.
A business decision, not an SEO one. Record it so it stops appearing as a task every month.
A citation count tells you a page matters. It does not tell you the page is available. Both facts are useful and they are different facts.
That is worth sitting with, because it is a more common situation than the tidy one. Some pages AI assistants trust are open to you. Some are editorially closed. Some you have chosen to stay off for reasons that have nothing to do with search. A citation count tells you a page matters. It does not tell you the page is available.
Both facts are useful and they are different facts. What the count did tell me is that aggregator-type pages carry real weight in my market, and the two I am on are tied to memberships I maintain rather than to anyone's editorial taste, which makes them the durable ones. Model preferences move. Association profiles do not.
And a competitor beating me where it counts
Broad discovery, cited rate of 92 responses
The plain "find me an advisor in Nashville" questions, September 2026. Six points, on the question type with the most people asking it.
In broad discovery, one competing Nashville firm was cited in 60% of 92 responses. I was cited in 54%.
Six points. I would never have known. There is no interface anywhere that shows an advisor that number, which is most of the reason I built the one that does.
The uncomfortable part of my own 57%
My citations lean heavily on a single page. Of all the citations the assistants made to my site, most of them point at my homepage, with my financial planning page a distant second and everything else in the noise.
Page-level citations to melbywealth.com, September 2026: 196 of 290 point at the homepage. This counts citations per page rather than responses, which is why the total does not match the 211-response figure above.
That is a fragile shape. A site whose AI citations rest on one page is one algorithm preference away from a much worse month, and it tells me my service pages are not doing the work my niche pages are. The 98% niche number and the 45% service number sitting in the same report is the same finding said twice.
I would not have known that either. It is the kind of thing that only appears when you count page by page instead of celebrating a site-level rate.
The other thing this tells me is, while content creation is important, what truly drives SEO and AI Search is focusing initial efforts on your main home page.
This is now a standing feature, not a one-time exercise
Everything above took most of a month to build the first time. It now runs by itself, on any advisor's firm, as the AI Citation Monitor inside AdvisorSEO Max.
You confirm your metro and your firm name. The system builds your question set, sends it to Claude, ChatGPT, and Perplexity with your city passed as the asker's location, scores every answer, and publishes your first report within the hour. It re-runs on the 15th of every month without being asked, with an on-demand run available in between when you want to check something you just changed.
What lands is the report above: cited and named rates separately with their sample sizes, per assistant with the model version attached, per question type with broad discovery on top, every page the assistants cited with your listing status on each, the specific questions competitors win, and a summary with next actions. It arrives as a PDF as well as on the dashboard, which matters more than it sounds like, because the PDF is what you forward to a partner or hand to a compliance reviewer without rewriting anything.
What it changes for an advisory firm
It puts a number on a channel that had none. Every advisor I talk to suspects prospects are asking AI assistants about them and has no way to tell whether it is going well. In my own practice that stopped being hypothetical this year: AI answers were the second-largest source of new prospects in 2026 through August, behind Google. A channel that produces prospects and no data is one you cannot manage.
It separates the work you control from the work you do not. The report distinguishes pages you own from pages you appear on from pages you cannot get onto. That sorting is the difference between a task list and a wish list.
It shows you the shape of your visibility, not just the level. A 57% rate built on one page and a 57% rate spread across six pages are different businesses, and only one of them is durable.
It closes the loop. Make the change, then read the next cycle's rate on the same question set. That is the difference between doing the work and knowing whether the work did anything. The same principle runs through the rest of the platform, where implemented recommendations get checked against real Search Console impressions and average position rather than assumed to have worked.
It is included in the plan at $199 a month with everything else, with no separate tier, because a measurement you only take when you have paid extra for it is not a measurement you will take every month.
What it will not do is promise you a rate. Nobody can, mine included. It measures, it points at the gaps, and the rate is the scoreboard.
What I would tell an advisor reading this
The ceiling works differently than Google. A results page shows ten links and there is a page two. An AI answer names three to six firms and stops. The question stops being where do I rank and becomes am I in the answer at all.
Judge yourself on the broad number. Niche and brand questions flatter everyone. The plain "an advisor in my city" question is the one with volume behind it.
Then check the shape. Count page by page. If one page is carrying your whole rate, you have a concentration problem your site-level number is hiding.
Sort the cited pages by what you can act on. Some listings you can join in an afternoon, some are editorially closed, and some you will choose to stay off for reasons unrelated to search. Do the joinable ones, then spend the rest of your effort on your own pages, which is the only part of the list nobody can revoke.
Be careful what you do with these figures. They measure visibility, not quality, and none of them belong in your marketing as a performance claim. I can say my site was cited in 57% of 368 answers in September 2026 because that is a first-party observation over a stated period. Nobody can tell you what your own rate will be. Anything you publish about your own visibility should go past your compliance reviewer first.
The honest caveats
Answers vary run to run, which is why every figure above carries its denominator rather than standing on its own. Of 384 requests, 368 came back with an answer, and the rates are calculated on the 368.
A measurement like this is only worth anything if it stays consistent: the same questions, the same location setting, the same scoring, cycle after cycle. Change the question set and you have started over.
And one thing worth saying out loud, because it happened here: the first two versions of this report disagreed with each other. Two different counting problems and an unordered database read were producing a different number every time the same data was scored. That is the failure mode of measurement software, and it is the reason a rate should be reproducible before anyone builds a decision on it. The numbers above come from a run that produces the same result twice.
Questions advisors ask
Is the AI Citation Monitor included in the plan or is it an add-on? Included. AdvisorSEO Max is one plan at $199 a month with no feature gates, and the monitor is part of it along with the Form ADV tools that shipped in the same release.
How long before I see my first report? The baseline cycle runs when you turn the monitor on and publishes within the hour. Monthly cycles run on the 15th after that, plus one on-demand run a month for checking something between cycles.
Do I have to do anything each month? No. The cycle runs whether or not you log in and the PDF arrives by email. The work is in the tasks the report surfaces, not in running it.
Can it check Google AI Overviews? There is no direct check for Google AI Overviews, and any tool claiming one deserves a second look. AdvisorSEO Max scores the structural foundation Google AI Overviews reads, which is the same schema, header, and firm-consistency work that helps everywhere else.
Does a low rate mean my website is the problem? Not always. In my own first month the largest single external gap was a directory listing rather than anything on my site. The report separates pages you own from pages you appear on so you know which one you are looking at.
Shaun Melby, CFP® is the creator of AdvisorSEO Max, SEO software for financial advisors and RIAs, and the founder of Melby Wealth Management, a fee-only RIA in Nashville. He runs every tactic on this blog on his own firm first. This content is educational and does not constitute investment, legal, or compliance advice. Consult your compliance officer before implementing any marketing changes.
See what to fix first on your own site
AdvisorSEO Max reads your Search Console data, ranks the fixes that matter, writes the replacement copy, and measures every month whether ChatGPT, Claude, and Perplexity cite your firm. Built by a practicing RIA founder, run on his own firm first.
Start your 14-day free trial Or see how the AI Citation Monitor works14-day free trial. Cancel anytime. $199 a month after.