TL;DR
An AI citation audit is the new quarterly review every marketing leader should be running. It is not a brand audit. It is not a search audit. It is the practice of asking AI tools the questions your prospects are already asking them, recording what comes back, and treating the results as a measurable input to next quarter’s plan. Below is the exact audit I run for clients in financial services, in five steps you can copy this week. No new vendor required. About six hours per quarter.
I get asked some version of this question every week. “How do we know if AI tools are recommending us?”
The honest answer is that you do not know unless you check. And most marketing leaders do not check. Which is strange, because the same leaders would never accept not knowing where they ranked on Google for their top ten keywords. The instinct to measure has not yet attached to the AI surface, and it needs to.
So here is the audit. I run it for clients every quarter. I run a lighter version for myself every month. It does not require a new tool, a new vendor, or a budget line. It requires six hours of focused work and a willingness to look at uncomfortable answers.
Step one: write the questions a real prospect would ask
This step is the one most people skip, which is why most audits produce noise.
Sit down with a blank document. Write the twenty questions a prospect would type into an AI tool if they were starting from zero. Not the questions you wish they would ask. The questions they actually do ask.
For a credit union, that looks like “what is the best credit union in [city],” “which credit unions offer the highest savings rate this month,” “is [your name] a good credit union for small business owners,” and seventeen more. For a community bank, that looks like questions about mortgage rates, branch locations, business banking, and trust. For a fintech, it looks like questions about features, pricing, integrations, and trust signals.
The twenty questions are your audit instrument. They will get reused next quarter. They are the most valuable artifact this audit produces.
Step two: run the questions through five AI tools, the same way every time
Pick five tools. The five I run are ChatGPT, Claude, Perplexity, Gemini, and Copilot. Use the free public version of each. Run the questions through each tool, copying and pasting the response into a tracking sheet.
Do not log in to a personal account that will color the answers. Do not run the audit in incognito and a normal window and average them. Pick one mode. Stick with it. The discipline matters more than the configuration.
What you want to capture for each question, in each tool: who got named in the response, what they got named for, what the response said about them in plain language, and whether any link was provided. The fourth column is the most important and the most overlooked. AI tools are increasingly providing source links. Whether your site is in those links is the closest analog to an organic search ranking that exists for AI surfaces today.
Step three: code the responses into three categories
Once you have the responses, code each one into three buckets. Named first. Named at all. Not named.
Named first means your institution was the first one mentioned in the response. This is the AI equivalent of a top organic result, and it is where you want to be.
Named at all means your institution was mentioned somewhere in the response, possibly third or fourth, possibly in a list of “other options.” Better than nothing. Worse than first.
Not named means you were absent entirely. This is the result you want to find, because it tells you which questions you have not earned a citation for yet.
Three colors. One spreadsheet. Hundred responses in total. The pattern shows up by the time you have coded the first thirty.
Step four: map the not-named questions to a content gap
Now the audit becomes a brief.
For each question where you were not named, look at who was named. Pull their content. Look at the page that the AI cited. Read it carefully. Three things will jump out.
The page will be more specific than yours. The page will name the product, the rate, the term, the location. The page will be structured in a way that an AI tool can lift cleanly. Headers that match the question. Answers in the first paragraph. Numbers in plain text, not in an image.
These are not new SEO principles. They are old SEO principles plus a new urgency. AI tools reward specificity in a way that Google search did not. A page that says “we offer competitive rates” will not be cited. A page that says “the rate on our 12 month CD is 4.85 percent, the rate on our 24 month CD is 4.65 percent, and here are the four other terms we offer” will be cited.
Build a content plan from this list. One page per missing citation. Half a day per page. Six pages a quarter is a reasonable pace.
Step five: rerun the audit next quarter
The audit is only useful if it becomes a series. The first one is a snapshot. The fourth one is a trend line.
Track the percentage of named-first responses quarter over quarter. Track the percentage of citations that go to your site. Track the named-at-all rate as a leading indicator before named-first moves.
Three numbers. One slide in your quarterly business review. The slide that finally answers the question every CEO is asking and every CMO has been struggling to answer.
What the audit will not tell you
The audit will not tell you why an AI tool recommended a competitor. It will not tell you which tools your members are using most. It will not tell you the conversion impact of an AI citation.
These are real limits. They are also second-order problems. You cannot optimize for AI citation rate until you can measure it. The audit is the measurement instrument. The rest of the work follows.
The bottom line
The marketing leaders I work with who started running this audit a year ago are now operating from a quarterly trend line. The ones who did not are operating from anecdotes.
Six hours per quarter. Twenty questions. Five tools. Three categories. One spreadsheet. Repeat.
The AI surface is not going to wait for a vendor to solve the measurement problem. The leaders who build the measurement themselves are going to lead the conversation in their organizations. The leaders who wait are going to keep being asked the question they cannot answer.
Run the audit. Then run it again.
Kevin Farley writes about AI visibility, AI readiness, and strategic growth for financial services. Read more on the blog.