You can track AI search leads in GA4 without a perfect referrer by combining the signals GA4 can observe with information collected when someone becomes a lead. Start by identifying visits with a recognizable AI source, record completed forms and other meaningful website actions, ask customers how they found you, and carry those answers into your lead records. Keep confirmed AI referrals separate from leads who report AI influence and from visits whose origin remains unknown.

A missing referrer is not proof that a visitor came from ChatGPT, Gemini, Claude, or Perplexity. Someone may see your company in an AI answer, search your name later, and submit an estimate request from a visit GA4 attributes to organic search. Someone else may click through from an assistant without a usable referring URL. GA4 measures the website activity it receives, not every recommendation that shaped the decision.

At Rank For AI Search, we recommend treating attribution as a chain of evidence rather than a hunt for one perfect traffic label: AI recommendation, identifiable visit, lead action, customer-reported discovery, and sales outcome. That distinction matters to home improvement companies because a roofing inquiry is not yet a booked inspection, and a booked inspection is not yet a sold job.

What counts as an AI search lead?

An AI search lead is an inquiry for which there is evidence that an AI-generated answer helped the customer discover, evaluate, or choose your business. The evidence may be a recorded visit from an AI platform, a customer saying an assistant recommended you, or both. A brand mention in an AI answer is valuable visibility, but it is not itself a lead.

We use four practical labels when discussing AI search attribution:

Observed AI referral: GA4 records a source associated with an AI platform, or a visitor arrives through a clearly tagged link that you control.
Customer-reported AI influence: A person names an AI assistant or AI-generated answer on a form or during a call.
Possible AI influence: The journey is consistent with AI discovery, but neither the visit nor the customer provides direct evidence. Keep this as a research category, not an attributed lead total.
Unknown origin: There is insufficient evidence to identify how the customer first found you.

These labels describe different types of evidence. One person may have both an observed referral and a customer-reported answer, so deduplicate by lead record before adding totals. An observed referral also identifies a visit, not necessarily the first time that person encountered your brand.

Why a perfect AI referrer is unlikely

A referring URL helps analytics identify where a website visit began. That signal may be unavailable when a person moves between an app and a browser, opens a copied link, uses privacy settings, or returns to your website another way. AI influence can also happen without any click at all. A homeowner might ask an assistant whom to call about water entering a basement, note your company name, and phone you later.

None of those paths produces a reliable rule such as “Direct equals AI.” Direct traffic can include typed addresses, bookmarks, untagged links, and other visits for which GA4 lacks source information. A visit to a detailed service page might deserve investigation, but its landing page and engagement do not establish which assistant, if any, sent the visitor.

There is a second distinction worth protecting: an AI crawler request is not a prospective customer visit. Crawler logs can help you examine access to important pages, but they do not show that a homeowner saw a recommendation or requested an estimate. Our AI search lead attribution setup guide covers the broader measurement problem; the steps below focus on what to configure and review in GA4 when referral data is incomplete.

Set up GA4 around leads, not just traffic

Step 1: Inspect the AI sources GA4 already receives

In GA4, open the traffic acquisition report and review Session source / medium. Search for the names and domains of AI services relevant to your customers, then inspect the landing pages and lead actions associated with those sources. Check the actual values in your property before writing a classification rule. A source name observed in one account is not a guarantee that every assistant click will appear under that name in yours.

Record a baseline for each recognizable source: sessions, landing pages, relevant lead events, and the date range reviewed. If a platform appears as Referral, that tells you how GA4 classified the recorded session. It does not tell you how many people saw your brand in that platform and never clicked.

Review these dimensions at the same time:

Session source / medium to identify recorded visit origins.
Landing page to see whether people arrive on a service, location, or educational page.
Event name to find completed lead actions rather than stopping at sessions.
Device category when testing whether phone-click behavior differs from desktop form submissions.

Start with source values you can inspect, not an exhaustive list copied from another website. AI products and referral formats can change, and a broad match may accidentally include unrelated sources.

Step 2: Create a reporting bucket for identifiable AI traffic

A GA4 custom channel group can make recognizable AI sources easier to review in one place. Build a channel whose conditions match the source values you actually found, then check its ordering so a broader rule does not classify those visits first. Compare the result against the original Session source / medium values before using the new group in a management report.

The key limitation is simple: a channel group classifies available data; it cannot recover a missing referrer. Do not reclassify all Direct sessions as AI. If you change reporting rules, retain a note of the logic and its effective date so future comparisons remain understandable.

For smaller sites, a filtered GA4 Exploration based on known source values may be enough to begin. The purpose is not to produce a more impressive channel total. It is to give your team a repeatable view of visits that have observable evidence.

Step 3: Measure completed inquiries as key events

Configure a lead event when a form is successfully submitted or an appointment is actually booked. If your site has several forms, decide whether they represent the same outcome or different stages. An estimate request and a newsletter signup should not carry equal weight in a contractor lead report.

We would separate website actions into two groups:

Completed lead actions: successful estimate submissions, completed consultation requests, or confirmed online bookings. Mark appropriate completed actions as GA4 key events.
Intent signals: phone-link clicks, chat starts, form starts, and visits to contact pages. Keep these available for diagnosis, but do not assume each represents a new lead.

A phone-link click only shows that someone tapped a link. It does not prove a connected call, a conversation, or a qualified project. If phone inquiries are important to your business, reconcile click events with your call records and ask your call team how each caller found you.

Test each event with a real submission or controlled test in GA4 DebugView. Confirm that it fires once, fires on success rather than merely on a button press, and appears with the expected service or location information where your implementation includes those fields. If a booking system sends customers to another domain, test the complete journey rather than assuming the event survives the handoff.

Step 4: Ask the customer, then preserve the answer

Add a short discovery question to estimate forms and call scripts: “How did you first hear about us?” Include an AI assistant option alongside search engines, referrals, social media, and other channels your business uses. If the person selects AI, an optional follow-up such as “Which assistant or answer do you remember?” can help you distinguish a named recommendation from a general impression.

Do not ask “Did you find us on ChatGPT?” as the only discovery question. It leads the customer toward one answer and overlooks other platforms. Likewise, avoid forcing a single source when the customer describes a multi-step journey. Someone may say, “An AI answer mentioned you, then I looked up your reviews.” Preserve both parts in the lead record.

Store the response with the inquiry, not just in a form notification that disappears from the workflow. Useful fields include the customer’s stated first discovery source, any named AI platform, the service requested, the date, and whether the inquiry became an appointment or sale. If you use a lead ID to reconcile records, use an internal, non-personal identifier. Do not send names, email addresses, phone numbers, or free-text responses containing personal details into GA4 event parameters.

Step 5: Connect website events to qualified outcomes

GA4 can tell you that a tracked action occurred during a recorded website journey. Your lead or job records tell you whether the inquiry was within your service area, matched a service you offer, received an appointment, and became a customer. Those are different questions and should appear as different measures.

For example, imagine a homeowner sees a waterproofing company in an AI answer, searches its brand name later, and submits an estimate form. GA4 may show an organic search session and a completed lead event. The customer may report the earlier AI recommendation on the form. Keep the observed GA4 source and the customer-reported discovery source as separate fields. Overwriting one with the other would erase useful information.

If your systems support an internal lead ID across the form submission and sales record, use it to reconcile the two without duplicating leads. Otherwise, establish a consistent manual review process. Either approach is more defensible than claiming a platform generated revenue because its name appeared in a referral report.

Build a report that separates evidence from inference

Once the collection process works, create a GA4 Exploration or recurring report for observed AI traffic. Pair it with a lead-record summary for customer-reported AI discovery and sales outcomes. A combined management view is useful, but each measure should retain its source and definition.

Measure	Where to check it	What it means
Recognizable AI-source sessions	GA4 source data	Recorded website visits with an identifiable origin
Lead key events from those sessions	GA4 events and source data	Completed tracked actions associated with those visits
Customer-reported AI leads	Forms, call records, or lead records	People who say an AI answer influenced discovery
Qualified AI-influenced inquiries	Sales or job records	Attributed inquiries that meet your business criteria
AI mentions and citations	Prompt and citation monitoring	Visibility in AI answers, not website leads
Unknown-source leads	Lead records	Inquiries without enough evidence to assign a discovery source

Deduplicate observed and customer-reported leads before showing a combined total. If the same homeowner clicked from an assistant and selected AI on the form, that is one lead with two supporting signals. Keep an uncertainty note beside any estimate of AI influence rather than folding possible cases into confirmed results.

Use the report to ask operational questions: Which service pages receive identifiable AI visits? Do those visitors complete estimate forms? Are callers naming AI platforms that never appear in GA4? Are the resulting projects in your service area? The answer may lead you to improve a page, change a form question, or repair a tracking event. It should not automatically lead you to credit every rise in Direct traffic to AI.

Where prompt monitoring fits, and where it does not

GA4 cannot show that an assistant recommended your company to a person who never visited your site. Prompt monitoring addresses that visibility question. Test consistent, relevant buyer questions and note whether your company appears, which services are described, which competitors appear, and whether the answer cites a source.

For a home improvement company, prompts should reflect real hiring decisions, such as choosing a roof replacement contractor in a city you serve or comparing approaches to a basement water problem. Record the prompt wording, platform, and test date. AI answers can vary, so one response should not be treated as a permanent ranking.

Rank For AI Search combines AI visibility audits, monthly monitoring and reporting, and, in its Growth and Market Leader programs, prompt tracking and competitor reporting. That work helps identify the questions on which a contractor is being recommended or overlooked. GA4 and lead records then answer a different question: what happened when people reached the business. For more on choosing meaningful visibility measures, see our guide to AI citation tracking metrics.

Best practices when the source is missing

Keep Direct traffic in its own category

Review Direct sessions that reach important service and city pages, especially when a lead action follows, but label the analysis as investigation. Deep landing pages, time on page, device type, and phone clicks can suggest where to look next. None proves an AI origin. Customer answers and identifiable source data carry stronger attribution evidence.

Tag links you control without mislabeling the channel

Consistent campaign parameters help you distinguish traffic from links you publish in emails, social posts, and other placements you manage. Name the actual source of that placement. An email link is still an email link even if the message discusses AI search. Do not apply an AI source label to owned links solely to make AI traffic appear larger, and do not assume you can add tracking parameters to links generated by independent assistants.

Check tracking after site or form changes

New landing pages, revised contact forms, changed confirmation flows, and external scheduling steps can disrupt lead-event measurement. Retest the journey after changes. If the form produces a lead but GA4 does not record its key event, an apparent drop in conversions may be a tracking issue rather than a drop in demand.

Separate crawler access from customer attribution

If important pages are inaccessible to relevant crawlers, that is a visibility issue worth diagnosing. It is not an explanation for how a particular lead arrived. Our article on AI crawler log analysis explains what server requests can help you investigate. Keep those technical findings distinct from GA4 sessions and customer inquiries.

Use the same definitions every month

Write down what qualifies as an observed AI referral, a customer-reported AI lead, and a qualified opportunity. Note when you add a new source rule or form answer. Consistency makes month-to-month changes more informative and prevents a reporting change from being mistaken for a sudden change in lead volume.

A real contractor perspective on measuring the whole pipeline

Steve Custus, CEO of Jamison Waterproofing, describes scaling his business from under $1 million to over $6 million within four years through a system of consistent leads, better appointments, higher close rates, and stronger margins while working with Daniel F. Fitzgerald III and his programs. He says the company is now positioning itself to be recommended through Rank For AI Search.

That account is about the broader lead and sales pipeline, not a measured GA4 result from AI search. It illustrates why we would not stop a contractor’s report at citations or website visits. If an AI-related inquiry is not followed through appointment quality and sales outcome, the business cannot tell whether that visibility is attracting the right projects.

Frequently asked questions

Can GA4 tell me exactly how many leads came from ChatGPT?

GA4 can report tracked lead actions associated with visits it identifies as coming from a ChatGPT-related source. It cannot count every person who saw a ChatGPT recommendation, contacted you without a website visit, or returned through another channel. Combine its observed data with customer-reported discovery for a fuller view.

Does Direct traffic include AI referrals?

It can include visits whose original source was an AI assistant, but it includes many other paths as well. Do not rename Direct as AI. Ask customers how they found you and keep unknown origins separate.

Should I create an AI Search channel group in GA4?

Yes, if you have recognizable AI source values and want a consistent way to analyze them. Match the values in your property, check rule ordering, and retain access to the underlying source and medium data. A channel group organizes recorded visits; it does not restore a missing source.

Is a phone click a lead?

Not necessarily. A click shows intent to call, not whether the call connected or concerned a valid project. Count completed conversations or qualified inquiries using your call and lead records, and use the GA4 click event as supporting website behavior.

Can I use server logs to identify AI-generated leads?

Logs can help diagnose requests to your website and technical crawler access. They do not establish that a particular homeowner saw an AI answer or became a customer. Use logs for technical investigation, GA4 for tracked website behavior, and customer and sales records for lead attribution and outcomes.

What if a customer says AI recommended us, but GA4 says Organic Search?

Keep both observations. Organic Search describes the recorded visit, while the customer’s answer describes an earlier discovery or influence point. Recording both gives your team a more accurate account of the journey than replacing one with the other.

What is the first step for a contractor with limited analytics resources?

Verify that completed estimate forms are recorded correctly, then add one neutral discovery question to forms and call scripts. Review the answers alongside GA4 source data and qualified lead records. A small, consistent process is more useful than a complex dashboard built on unreliable events.

Make AI lead measurement useful, not artificially precise

The strongest GA4 setup does not promise to uncover every hidden AI referral. It records observable visits and completed actions accurately, preserves what customers say about their discovery journey, and follows inquiries through qualification and sales. That gives a home improvement business a defensible way to judge AI search without treating unknown traffic as proof.

We help contractors improve the visibility side of that equation through AI footprint audits, structured and readable content, off-site brand alignment, prompt tracking, and ongoing reporting. If you want to examine where your company is being recommended and how those recommendations fit your lead pipeline, book a consultation with Rank For AI Search.