CallRail says AI answers bring 1 to 2 percent of tracked calls

Sean McCrohan, VP of Technology at the call-tracking company CallRail, told Search Engine Journal that clicks from AI answer engines now account for about 1 to 2 percent of tracked calls across its customer base, and that the figure has roughly doubled since January. The article, by CallRail's McCrohan and Steve Wiideman with Loren Baker of Foundation Digital, ran on Search Engine Journal on 14 September and gathers the first practical numbers on turning AI search visibility into local phone calls.

The numbers CallRail is seeing

McCrohan reports two patterns from CallRail's own data. The first is size: AI citation clicks are 1 to 2 percent of calls for the customers CallRail measures, up roughly two times since the start of the year. The second is timing. In classical Google search, about half of website traffic on the sites CallRail tracks arrives outside regular business hours; for AI-referred sessions, that share rises to nearly two thirds.

Both numbers are CallRail's own, from calls they attribute. They are small, and McCrohan says so directly. What they suggest is that AI traffic is meaningful enough to measure and different enough in pattern that a business tracking it with the same reporting it uses for organic Google will read it wrong.

Where the AI answer disagrees with the local pack

The article also cites analysis by Steady Demand on repeat local queries. Running the same local search inside Google's classical local pack returns the same top business about 90 percent of the time, while running it inside Gemini returns the same top business only around 7 percent of the time. That gap says a chosen brand is far less stable inside an answer engine than inside a map result, and it means one placement in Google's local pack has a different value from one placement inside an AI answer.

What the article tells operators to change

The article gives five techniques, aimed at operators who already run multiple locations.

Phone numbers get swapped server-side rather than in the browser, so an AI crawler sees the tracking number and the human visitor sees the one the business publishes. Prompt libraries carry about 100 to 125 short claim sentences that state what the business does in the customer's words, tested against the model that answers a customer. Call recording transcripts get read for the phrases customers actually use, and the website copy gets updated to use the phrases it was missing. Reviews get collected on Reddit, TripAdvisor and Yelp as well as Google, with a target average around 4.5 to 4.7, on the reasoning that answer engines draw on a wider set of sources than the map pack does. And schema, feeds and tracked numbers stay under one owner across locations, so a franchisee cannot open a listing that sends data to a third party.

For a team building marketing software, the honest summary is that measurable AI traffic is still small, and phone calls remain the most reliable measure because a click through an AI citation carries a referrer and a call does not. A business that treats AI-referred sessions as a separate group, with its own time-of-day baseline and its own attribution, will see the growth before it shows up in weekly totals.

Source

Primary source: How To Connect AI Search Visibility To Local Leads, by Sean McCrohan, Steve Wiideman and Loren Baker, Search Engine Journal, 14 September 2026.