AI search visibility for advisory firms
How to check what AI says about your firm
How to measure what AI says about your firm by hand: the questions to ask, how many times, what to record for each answer and the reports to read.
Gino FerrandUpdated 10 min read
Ask the questions your clients would ask, in each AI engine, more than once, and write down what every answer says about you. The repetition is the point, because answers to the same question change from run to run, so any single answer shows you one moment and little more. What counts is how often your firm is named, recommended, cited and described correctly across many runs, read next to the reports from Google, Bing and Google Analytics.
45.5% of the pages cited in Google’s AI Overviews were new from one check to the next, on average, across more than 43,000 keywords tracked for a month.
The check, step by step
The check runs in a fixed order, from the questions your clients would ask to the pages you fix, and you repeat it on a schedule. The order matters because you cannot score answers you never wrote down, and you cannot choose which source to fix until the record shows which pages the answers drew on. That makes the record in the middle the step that turns a handful of answers into something you can compare month to month.
One check proves little because the answers move
In the studies below, which tested brand and product questions and none about advisory firms, the same question asked on two different days almost never returned the same list. Glen Allsopp at Detailed ran more than 1,300 commercial prompts every day for 28 days. In both ChatGPT and Google’s AI Overviews, the full list of brands for a prompt changed almost every day.
What held steady was the brand named most often for each prompt, which still appeared on most days in both engines. The leading brand tends to stay while the rest of the list changes around it, which is why your check counts how often your firm is named.
So the number worth keeping is the share of a question’s runs that named you. Rand Fishkin of SparkToro, whose study put the chance of two runs giving the same list of brands at less than 1 in 100, came to the same view. In his words, “I’ve been swayed from my initial position and now believe visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric.” He added, “But, any tool that gives a ‘ranking position in AI’ is full of baloney.”
Four studies point the same way. The last of them, from Growth Memo, also shows that the engines differ from each other, which is why your check should cover every engine your clients use.
| What they ran | What they found | What it means for your check | |
|---|---|---|---|
| SparkToro, January 2026 | 2,961 runs of 12 prompts in ChatGPT, Claude and Google’s AI, by 600 volunteers | Less than a 1 in 100 chance that two runs give the same list of brands | Run each question more than once |
| Detailed, September 2026 | 1,300+ commercial prompts daily for 28 days | The full list changed almost every day; the most-named brand stayed on most days | Count the days you are named |
| Ahrefs, November 2025 | 43,000+ keywords for a month | Nearly half the cited pages were new at each check, while the answers’ meaning stayed about the same | Record the cited pages every run |
| Growth Memo, May 2026 | About 20,000 prompts in ChatGPT, Perplexity and AI Overviews | 91.07% of cited pages appeared in one engine only | Check every engine your clients use |
Source: SparkToro, Detailed, Ahrefs and Growth Memo (Kevin Indig), read November 14, 2026
The questions to ask
Ask the questions a prospect types before choosing a firm like yours, in their own words. Write them down once and keep the wording fixed, so that when the answers change, you know the engine changed them. Mix questions that never mention you with questions that do, because the first kind shows whether an engine names your firm unprompted and the second shows what it says about you once your name is in the question.
- Who you serve. “Which financial advisors in Denver work with physicians?”
- The problem you solve. “Who can help me plan for taxes after selling my company?”
- Your firm by name. “What does Your firm do, and who is it for?” Check every fact in the answer.
- The comparison. “Should a business owner choose Your firm or Competitor 1?”
Run each one in ChatGPT, Gemini, Claude, Perplexity and Google’s AI Mode. Use the same account settings each time, and note them, because memory in a signed-in account and your location can both shape the answer. The work adds up quickly, since twenty questions in five engines on two days comes to 200 answers a month.
If that is more than you can manage, start with five questions in the three engines your clients use most, on two days, which makes 30 answers, and add questions once the habit holds.
What should you record for each answer?
For each answer, record whether your firm is named, recommended, cited and described accurately, and keep those four results apart, because one answer can name you without choosing you. Add the date, the engine, the exact question and every page the answer drew on. When you run the Denver question from your list, one engine might answer like this.
Which financial advisors in Denver work with physicians?
- Competitor 1 RecommendedPlans for physicians, cited from its own page on residents’ loans.
- Your firm NamedListed from a directory profile that gives an old address.
- Competitor 2 NamedGeneral planning; no source shown.
Your firm makes that list, but Competitor 1 gets the recommendation, and the one fact the answer gives about you, an old address from a directory profile, is wrong. So the fix belongs on that directory profile, because that is where the engine found the address. A record sheet keeps each of those results visible.
| What to write | Why | |
|---|---|---|
| Date and engine | The day, the engine and whether you were signed in | Answers change by day and setting |
| Question | The exact words, unchanged from the last run | So runs can be compared |
| Named | Yes or no | The broadest result |
| Recommended | Yes if the answer picks you or calls you the best fit | Being named is not being chosen |
| Cited | The link to your page, or to a page about you | Shows which page the engine trusts |
| Accurate | Each wrong fact, word for word | A wrong fact needs a fix at its source |
| Sources | Every page the answer linked | Where to work next |
| Others named | The other firms in the answer | Who you are compared with |
Then score each result as a share of runs, the measure the studies above point to. Count how many runs of a question named you out of all the times you asked it. Six runs of the Denver question in one month might leave Your firm with a tally like this one.
| One question, six runs | Your firm | What it points to |
|---|---|---|
| Named | 4 of 6 | Two runs left you out |
| Recommended | 1 of 6 | Named, but rarely the pick |
| Cited | 1 of 6 | Your own pages are rarely the source |
| Accurate | 2 of the 4 that named you | A wrong fact to fix at its source |
On that tally Your firm looks healthy on the first line, named in four runs of six, yet only one run recommended it, and two of the four answers that named it got a fact wrong. Compare those shares from one month to the next, because a single answer moves too much to tell you much on its own.
Qubo’s free AI Visibility Audit runs this check for your firm and scores it the same way. It is where our AEO work starts.
Which reports from Google and Bing show AI visibility?
Google Search Console, Bing Webmaster Tools and Google Analytics each show part of your firm’s AI visibility, and no two of them see the same part. Search Console counts how often you appear. Since August 31, 2026, it shows every site its impressions in AI features such as AI Overviews and AI Mode, by page, country, device and date, although impressions are its only measure so far.
Bing Webmaster Tools looks at citations instead. Its AI Performance report, in public preview since February 2026, counts how often Bing’s AI answers cite your pages and for which grounding queries, and since June 2026 it also shows your share of all citations for each grounding query. Google Analytics covers the visits. Since May 13, 2026, it sorts visits from assistants such as ChatGPT, Gemini and Copilot into an AI Assistant channel, while visits from Google’s own AI features still count as organic search.
None of them shows what ChatGPT, Claude or Perplexity say when they do not send a visit, which is why the hand check stays. Ask new clients how they found you as well, and give “an AI assistant” its own answer on your intake form.
How do you find what an answer was based on?
Open the sources an AI answer lists and note every page it drew on. When a wrong fact traces to a directory, an old article or a profile you no longer use, fix it there, then ask again on your next run. Our ChatGPT guide covers which outside pages matter most for advisory firms.
If no answer can reach your site at all, start with access. Our crawler self-check covers robots.txt, CDN settings and indexing in ten questions.
What a self-check cannot tell you
A self-check cannot tell you what your prospects actually type. Your questions are your best guess, and Bing’s grounding queries come closest, but what Bing reports are the queries it used to ground its answers, while the prompts people wrote stay out of view. Nor can the check show you the answer a particular prospect saw. Answers vary from run to run, and memory and location can shape them, so your record holds the range of what the engines say.
It also cannot tell you why a number moved. When a share rises after a fix, the cause may be the fix or a change in the engine, so give it several runs before you credit either.
Sources
- SparkToro (Rand Fishkin and Patrick O’Donnell), AIs are highly inconsistent when recommending brands, January 27, 2026.
- Detailed (Glen Allsopp), How volatile are AI responses?, September 23, 2026.
- Ahrefs (Louise Linehan), AI Overviews change every 2 days (but never change their mind), November 11, 2025.
- Growth Memo (Kevin Indig), The consensus gap, May 11, 2026.
- Google Search Central Blog, generative AI performance reports, June 3, 2026, updated August 31, 2026.
- Bing Webmaster Blog, AI Performance in Bing Webmaster Tools, February 10, 2026.
- Bing Search Blog, New AI visibility insights in Bing Webmaster Tools, June 16, 2026.
- Google Analytics Help, New AI Assistant traffic measurement, May 13, 2026, and Default channel group.
Questions
How often should a firm check what AI says about it?
Monthly suits most firms, with the same questions each time. Check again a few weeks after a big change, such as a new website or a corrected directory profile.
Are AI rank-tracking tools worth paying for?
Not if they report a single position, because answers change too often for one to mean much. A share of runs that name you, across many questions and runs, is the measure worth tracking.
Does Search Console show traffic from ChatGPT?
No. Its AI report covers Google’s AI Overviews and AI Mode. Visits that ChatGPT, Gemini or Copilot send show in Google Analytics, in the AI Assistant channel.
How do you measure whether AEO is working?
Track the share of runs that name, recommend, cite and correctly describe your firm, month to month, next to AI impressions in Search Console and inquiries from people who say an AI assistant sent them.