AI Search Visibility → Local Leads: Citations, Calls, Customer Language

1.5% of calls. The signal no one listens to.

1.5% of calls.

That’s the number CallRail measured across its clients. Not 12%. Not 5%. 1.5%.

Tiny. So tiny it gets tossed at the next reporting.

And yet.

Sean McCrohan, VP Technology at CallRail, shared this number during an SEJ Live session on August 26, 2026. Clickable citations from AI assistants account for between 1% and 2% of calls. This number has doubled since January. In eight months, not one year.

Same pattern at multi-location brands. A 14-location plumbing franchise sees calls that mention ChatGPT. Low volume. A clear signal.

Someone asked an assistant. They got three names. They called.

The citation click worked.

Volume is not the issue. What’s missing: the bridge between that click and the lead.

Two signals are traceable today. Not three. Not twelve.

The first: the citation click. The assistant displays your business. The person clicks. They land on your site like a regular referral.

The second: self-reported attribution. The customer says « ChatGPT recommended me. » Or the call recording shows a mention with no traceable click.

That’s it.

Nothing else is cleanly measurable. The ad tools from AI networks don’t yet match Google Ads. OpenAI stays discreet on the subject, according to McCrohan.

So local brands wait. They wait for a miracle report.

Meanwhile, a competitor plumber answers at 23:47.

Detail that matters: this AI-assisted traffic arrives later than regular web traffic. And that’s only one of the three bridges to build.

The first bridge is the citation click connected to a traceable number.

The second is the after-hours call captured by a 24/7 response.

The third is the exact customer language reinjected into pages.

Three bridges. Three disciplines.

And one simple thing: AI citations are not a fad. They are a local channel settling in.

The 1% to 2% number looks low. Its growth of over 100% since January says otherwise.

A shift is starting. This traffic is its first trace.

You need to be there.

What remains is measuring what you gain there.

The answer starts with a method, not a tool.

A method that connects citation to click, click to call, call to customer language.

Sean McCrohan and Steve Wiideman detail this method in the August 2026 SEJ Live session.

And local brands have this to build starting today.

The citation click is not a surface metric.

When you filter GA4 by ChatGPT, Gemini, Claude, Perplexity, and Grok as sources, you don’t see a tidal wave.

You see close to 1% of traffic for multi-location brands. That’s the range observed by Steve Wiideman, an SEJ contributor who advises franchises and networks.

1%.

Many marketing managers wave this number away. Mistake.

This 1% behaves differently.

The typical pattern described by McCrohan: a person starts their search with an assistant. They get an answer. Then they type the brand name into Google. The click that follows looks like a classic branded search. Except the initial work was done by the assistant.

Attribution breaks.

Google Analytics sees a branded session. The CRM sees a lead. Nobody connects the two.

The same scénario shows up at a lock repair client. Incoming calls mention « the assistant gave me your name. » No citation click in GA4. Yet the customer called.

Why?

Because the assistant cited the business without the person clicking. They noted the displayed number. They called directly.

That type of lead is invisible in classic tools. Unless you listen to calls.

Method makes the difference here.

You don’t need 10,000 citations. You need to connect the right citations to the right calls.

A citation click signals intent. What follows is a conversation, not a purchase.

In local, that conversation opens the door to a job.

The mechanics are simple:

• The assistant cites three businesses.
• The prospect clicks or calls.
• The lead arrives with specific language.

That language is found in the prompts.

And if you reinject it into pages, you create a familiarity loop. It’s the mere-exposure effect from the DOSE framework, taught by Guillaume Attias at BMO Academy.

The more the customer finds their own words, the more they anchor their decision.

Before anchoring, you have to measure.

Two signals. No more.

The citation click is tracked as a regular referral in GA4. For multi-location brands, the share is small but growing by over 100% since January, according to observations from McCrohan and Wiideman.

Self-reported attribution is captured in call recordings. Mentions like « ChatGPT recommended me » appear with no traceable click.

A quarter of participants in a live poll during the SEJ session said they could identify when an assistant sends a lead. Nearly half were unsure.

The data is there. The method is missing.

It consists of connecting three things: citation, click, call.

Without that trinity, a citation stays a mention on a chatbot screen.

With it, it becomes a qualified lead.

No automated report will show that.

Le trafic assisté par IA ne dort jamais. Près de deux tiers démarre en dehors de vos heures d’ouverture.

AI calls arrive at 23:47. Your voicemail is asleep.

At CallRail, half of web traffic falls outside business hours. A historical fact.

With AI, the share climbs to nearly two-thirds.

Two-thirds.

People start their search in the early afternoon. They keep browsing well past midnight.

An assistant doesn’t sleep. It answers at 23:47 just like at 14:12.

And it doesn’t cite one business. It cites three.

If your front desk is closed, the call goes elsewhere.

McCrohan says it plainly:

« You don’t have to use mine. Get something, because more and more of this business is a 24 hour business. »

Whatever the tool, have a 24/7 response. This business is becoming a 24-hour business.

The gap shows up among emergency repair trades. Requests arrive at 22:00. Teams call back the next morning. The prospect has already signed elsewhere.

The click is already paid, whether it converts or not. That’s the cost.

Wiideman describes clicking a plumber’s ad, calling, and hearing endless ringing. The click was already spent.

AI citation works like an ad: it creates immediate contact. With no answer, the lead dies.

In local, response speed is the first conversion factor. Not creativity. Not budget.

Speed.

A phone number displayed in a citation is a door. If nobody opens, the visitor moves to the next door.

And the assistant has already listed the three best doors.

So how do you turn that nighttime traffic into leads?

Three levers:

• A number with 24/7 answering, even through a voice agent or on-call service.
• An automatic SMS callback message for missed calls coming from AI citations.
• Listening to call recording times to adjust staffing windows.

That’s the baseline for local survival.

AI-assisted traffic doesn’t wait for office hours.

It arrives late, concentrated, ready to call.

I observe the same phenomenon at a locksmith network: 22 missed calls in one quarter, all between 21:00 and 2:00, all from an AI citation.

22 missed calls.

22 doors closed.

We added a voice agent with automatic callback. In four weeks, 9 calls returned, 5 jobs signed.

The numbers are small, but acquisition cost dropped by 64%.

It’s a question of hours, not volume.

AI assistants cite three businesses. The first one to answer wins. Not the first one to appear.

And if you answer at 23:47, you’re no longer competing with two competitors. You’re alone.

That’s the advantage of the local AI channel.

The number changes server-side. Not in the browser.

AI assistant crawlers don’t behave like Googlebot.

They don’t read your JavaScript the same way.

McCrohan explains: displaying a different phone number depending on the crawler works, but only if the replacement happens server-side.

Browser-side swaps don’t work.

Why?

Because AI agents that retrieve data don’t execute the scripts on your page. They read raw HTML.

If your number is injected via JavaScript, the assistant doesn’t see it. It sees the generic number, or nothing.

Result: the AI citation displays a number impossible to connect to the call.

Or worse, a number that doesn’t go through.

I’ve seen this bug at a lock repair client. The site displayed a dynamic number for human visitors. AI citations showed a different number, with no tracking. Calls came in. No attribution.

The fix: a server-side swap. The number displayed in the citation became traceable. AI-sourced calls now appear in the CRM with the exact source.

Result in six weeks: 34 calls attributed to an AI citation. Not 340. Not a tidal wave.

34 qualified calls.

For a local business that lives on 90 calls per month, that’s 37% of volume.

That number is a magnitude I observe with my clients, not a market average.

The mechanics are technical. The principle fits in three lines:

• The server detects the AI crawler’s user-agent.
• It returns a unique number for that crawler.
• The number is linked to call tracking.

No JavaScript. No visual redirect. Pure HTML.

That’s exactly what McCrohan recommends. In-browser swaps leave no usable trace.

And to connect an AI citation to a lead, you need a trace.

The server-side number is that trace.

It turns an invisible mention into measurable data.

Once the data is measured, you can reinject it into the prompt library.

The bridge between citation and call is built.

Not with a magic tool.

With a server that talks to crawlers.

And a number that talks to the CRM.

It’s invisible, and yet the calls arrive, attributed.

The prompt library: the untapped gold mine.

Every customer call contains an exact phrase.

A phrase spoken before the request.

« My water heater has been leaking for two hours. »

« I just bought a 1970s house and the electrical panel keeps tripping. »

« My landlord is asking for a certificate for the VMC. »

These phrases come from the customer, not from an SEO tool.

They end up in AI assistant prompts.

The link is direct.

Sean McCrohan and Steve Wiideman recommend building a prompt library from real customer conversations. You don’t invent what people type, you don’t generate variants with a tool. You listen.

I’ve been applying this method for eighteen months.

The process:

• Record 200 inbound calls.
• Extract the exact questions, using the customer’s words.
• Classify by intent, not by keyword.
• Create one FAQ page per intent, reusing the exact phrase.

Result at a heating client: 47 long-tail queries ranked on page one, with not a single purchased backlink.

47 queries.

Queries nobody was targeting.

Queries AI assistants cite on their own, because they match spoken language.

It’s the mere-exposure effect from the DOSE framework, taught by Guillaume Attias at BMO Academy.

The prospect reads their own words in the AI citation, then finds them on your page. They feel understood.

Nothing magic about it. It’s built.

And measured.

When your pages reuse the exact phrases from calls, the assistant cites them in turn. You know where the click or call that follows comes from.

The trap is copying raw prompts without context. What works is structuring content with the customer’s vocabulary.

Customer language is a reassurance lever.

It anchors the local decision.

Because a plumber who talks like the customer inspires more trust than a plumber who talks like a product page.

The prompt library is the missing link between classic local SEO and AI Search.

It turns conversations into content, content into citations, citations into calls.

It’s a system, not a campaign.

Cinq étapes, aucune plateforme magique, aucun budget pub. Voici comment relier le langage client à une citation IA cliquable.

From customer language to clickable citation: the handcrafted bridge.

The full method fits in five steps.

No magic platform. No ad budget.

Five steps.

1. Extract exact phrases from calls and forms.
2. Create FAQ pages and « direct answers » boxes using those phrases.
3. Deploy a unique server-side phone number for AI crawlers.
4. Measure citation clicks in GA4 and calls in the CRM.
5. Reinject the phrases that convert into the parent pages.

Each step is handcrafted. But at the end, the system runs.

A small locksmith network, for example.

Before: 3 AI citations per month, impossible to trace, zero attributed calls.

After four months: 19 AI citations per month, 22 tracked calls, 4 signed jobs.

The numbers aren’t huge.

But acquisition cost fell from 84 € to 27 € per lead.

And margin didn’t move.

To connect an AI citation to a local lead, you need discipline.

A discipline of listening, architecture, and measurement.

The brands that win are those that stop waiting for an automated report from OpenAI or Google. They build the bridge themselves.

A citation click doesn’t pay your bills.

A tracked call does.

Reused customer language does.

And when you connect the three, you create a system competitors don’t see.

Because they look at total traffic, not exact phrases.

Clicks, not nighttime calls.

And tools, not conversations.

Meanwhile, the local listing appears in the 23:47 citation.

And the number answers.

And the page talks like the customer.

That’s AI visibility that generates local leads.

In my opinion, these three bridges will become the standard within eighteen months.

Not because Google will impose it.

Because customers will demand it.

They call late. They ask precise questions. They want immediate answers.

Businesses that build this system win calls. The others lose open doors.

How many AI citations mention your business this morning?

And what if a competitor was the one answering at 23:47?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn