AI search is generating local business leads right now, but the numbers are easy to dismiss. CallRail data shared at SEJ Live puts AI citation clicks at roughly 1% to 2% of customer calls. That share has roughly doubled since January — a meaningful trajectory even if the raw volume is still low.
A small signal with two trackable paths
Sean McCrohan, CallRail’s VP of Technology, and local search advisor Steve Wiideman walked through what marketers can actually measure. McCrohan said a common journey starts with an AI assistant before someone does a branded search. As a result, the final tracked visit can look like a normal organic or branded session, which hides the AI assist.
Right now, two signals are trackable:
- Citation clicks: a user clicks a source inside an AI answer and lands on your site like a referral.
- Self-reported attribution: a customer says ChatGPT or another assistant recommended the business, even without a click.
Both are up more than 100% since January from a small base. In GA4, filtering for ChatGPT, Gemini, Claude, Perplexity and Grok produced close to 1% of traffic for the multi-location brands Wiideman advises — a discovery channel, not a primary sales channel.
AI demand arrives after hours
CallRail ordinarily sees about half of website traffic outside business hours. With AI search, McCrohan says that share rises to nearly two-thirds. Users may begin browsing in the early afternoon and continue past midnight. AI answers also tend to surface multiple businesses rather than one winner, so a missed call can push the lead to a competitor.
McCrohan told marketers to “get something” — his company’s voice agent or another option — because local AI discovery is becoming a 24-hour business.
What works for attribution
Server-side phone number swaps work for AI crawlers; in-browser swaps do not. McCrohan explained that AI agent crawlers do not execute page scripts, so browser-level changes go unseen. Keeping a consistent name, address and phone number remains critical, and Wiideman warned marketers to test user-agent tactics on low-traffic pages to avoid cloaking risk.
Build a prompt library as a trend line
Wiideman recommends starting with semantic triples: specific claims a business wants to be known for, turned into roughly 100 to 125 prompts. But don’t obsess over single answers. “Prompt drift” means the same query can return different results. Steady Demand analysis found that repeating a local search in Gemini produced overlapping cited sources only about 40% of the time, with the same top business in only about 7% of results — compared with about 90% for Google’s local pack.
He also advised local brands to focus on the type of site being cited, not one specific source.
Use the customer language you already have
Call transcripts capture the words customers actually use, which may differ from site copy. Chat logs are another overlooked source. Reviews still feed recommendations: Wiideman suggests training locations to collect ratings beyond Google Business Profile — including Yelp, which feeds Bing and Apple Maps and has a ChatGPT partnership — and to aim for roughly a 4.5 to 4.7 average.
Governance scales better than chasing signals
For multi-location brands, centralized control of schema, data feeds and tracked phone numbers prevents rogue listings and unauthorized numbers. Wiideman says consistent business data matters more than traditional ranking factors for multi-location AI searches.
Looking ahead, both speakers expect AI agents to verify a phone number and offer to place the call while the customer never visits the site. McCrohan described the shift as exciting but not entirely new, with enough carryover from existing SEO and local search work to keep teams grounded.
Source: Search Engine Journal
