Do AI Citations Actually Turn Into Calls? What We Measure, and What We Can and Cannot Attribute

Short answer: Sometimes, and the path is indirect. When an AI answer names a firm, the person usually does not click a source link (Pew measured that at about 1 percent of searches with an AI summary). They search the firm's name, or type the domain, or call the number the answer gave. The evidence therefore shows up as branded search, direct traffic, a small stream of referrals from chatgpt.com and perplexity.ai, and intake calls that say "ChatGPT recommended you". A firm that tracks those four together can see the effect. A firm that looks only for clicks from AI engines will conclude, wrongly, that there is none.

Why is attribution hard?

Because the surfaces that name the firm are designed to answer, not to send traffic. An AI Overview or a ChatGPT answer that says "firms in Pasadena handling this include Smith & Lee" has done its job; the reader may act on it without touching a link. Analytics sees the action later, disconnected from the answer. We have watched this in our own logs since 2024: a citation appears for a process question, and a week or two later the firm's branded impressions tick up, without a corresponding referral.

That is also why claims of "X leads from AI" from any vendor should be read with the method next to them.

What signals can be measured?

The intake question is the one most firms skip and the one that matters most. "I asked ChatGPT" is now a common answer in the practices where the demographic skews older and affluent, and it is invisible unless someone writes it down.

What does the pattern look like when it works?

Described in words, because the numbers are per firm and we do not publish client data:

1. A specific page (a courthouse guide, a cost page, a process page) begins to be cited in AI answers for the questions it answers. This shows in the monthly run first.
2. Two to six weeks later, branded search impressions rise above the previous baseline.
3. Referral sessions from AI domains appear in small numbers, landing on that page.
4. Intake records a few "found you through ChatGPT / Google's AI" answers.
5. Calls from the practice area that page serves increase, without a corresponding rise in organic clicks to it.

Step five is the one that pays. Steps one through four are how a firm knows step five was not a coincidence.

What can we not attribute?

• A specific call to a specific AI answer, in most cases. The chain is broken by design.
• The share of branded search that is AI-driven versus referral-driven. It is a mix.
• Anything from a single month. The signals are small and they lag.

What does this mean for what a firm pays for?

It means the measurement has to be built before the work, and the report has to show all four signals together with the citation rate, month after month, rather than a single "AI leads" number. It also means the work that produces citations (specific pages, entity consistency, reviews, directory accuracy) is the same work that produces map-pack calls, so a firm is not betting its budget on an unattributable channel; it is getting the second effect from the first spend.

What does not work?

• Counting AI referral sessions as the whole effect. They are the smallest part of it.
• Asking the model "would you recommend us" and calling a yes a lead.
• Vendors reporting "AI-attributed revenue" with no intake data. Ask how.

How should a firm start?

Add the intake question this week and record the answers verbatim. Filter branded queries in Search Console and note the baseline. Set up the referrer filter in analytics. Then run the citation measurement monthly, by the method in how to track your AI citation rate. The AI citation audit sets up all of it for a firm, and the free visibility check gives the starting citation rate. What the retainer covers, and how it is reported, is on the pricing page.