Track AI referrals using the source or campaign values actually recorded, then connect them to qualified inquiries and booked work. Save first-touch attribution separately from a customer's answer to how they found you. A mention, citation, visit, contact-button tap, and confirmed client are different measures. Manual prompt checks are dated observations, not a ChatGPT market-share score.
Keep the measures separate
A useful report distinguishes a business mention, a linked citation, a recorded visit, an inquiry, a qualified lead, and booked work. A contact-button tap shows intent; it does not prove a call connected. A form-submit event can show an attempt; confirm delivery and qualification before counting it as a lead.
This separation matters when deciding whether more content is earning business. Traffic can rise without better inquiries, and a person can contact you after a recommendation without clicking the assistant's link.
Find the referral sources you actually observe
In GA4, open Reports, Acquisition, Traffic acquisition and inspect Session source or Session source/medium. Look for observed AI-related sources such as chatgpt.com or perplexity.ai. Do not require a particular medium label or assume every assistant sends a referrer.
Google's traffic acquisition documentation explains how campaign parameters populate attribution. If an incoming URL contains utm_source=chatgpt.com, preserve that observed value. Do not assume every ChatGPT click uses that parameter. GA4's Page location retains URL details that are not shown in Landing page + query string.
Keep the first landing page and source, then inspect later inquiries. Exclude your own tests where possible and compare equivalent date ranges using the same measurement definitions.
Ask how the person found you
Add an optional discovery-source question to the contact process. Keep the answer in its own field, separate from recorded first-touch attribution. Someone can first arrive from Google and later report that ChatGPT helped them choose, or hear a recommendation and type the URL directly.
When a customer mentions an assistant, record the engine, approximate question, service, location, and the answer or source link if they are willing to share it. Do not request a private conversation transcript when the discovery details are enough. This record helps choose which existing page to improve next.
Use each platform's report for its own scope
Google's Search Console Generative AI performance report covers impressions in AI Overviews and AI Mode. A property may not show the report if it has insufficient AI impressions. It is not a ChatGPT referral report.
Bing's AI Performance announcement describes citations, cited pages, and grounding-query samples across Microsoft Copilot, Bing summaries, and selected partner experiences. Citation counts do not show traffic, rankings, or qualified leads, and the published scope should not be relabeled as ChatGPT performance.
Keep prompt checks reproducible
Use a small fixed set of relevant questions. Save the exact prompt, engine, date, location context, returned answer, source URLs, and any incorrect business facts. Keep linked citations separate from unlinked mentions.
Repeat with conditions as consistent as practical. Answers can vary by context and query rewriting. A change in the saved sample is a reason to investigate; it does not prove that one content edit caused the result or that you hold a percentage of all ChatGPT recommendations.
Close the loop on actual business
Review inquiries for service fit, location fit, and a genuine contact method. Record whether the conversation happened, a quote was sent, and work was booked. Use your own confirmed records for revenue, and keep test submissions out of the totals.
Report the observed referral cohort and the self-reported cohort separately, with a combined view that avoids counting the same inquiry twice. Note attribution gaps plainly. Our AIO service uses these observations to prioritize access, factual corrections, and useful content; it does not promise that a citation count will turn into a set number of clients.
Common questions
Can you track ChatGPT traffic?
You can track visits with observed ChatGPT referrers or campaign values and connect those records to later inquiries. Some visits or offline recommendations have no usable source, so also keep self-reported discovery.
Is a contact-button tap an AI lead?
No. It is an interaction event. Confirm the call or inquiry, delivery where relevant, and qualification before counting a lead. A booked client is a later outcome.
Should self-reported discovery replace first-touch attribution?
No. Save both separately. They answer different questions and can both be correct when a person discovers a business through one route and chooses it through another.
Does Search Console show ChatGPT citations?
Google's Generative AI performance report covers its AI Overviews and AI Mode impressions. It does not establish ChatGPT citations or referrals.
How should I measure manual AI tests?
Keep dated records of the exact prompt, engine, location context, answer, and sources. Treat a fixed prompt set as directional observations rather than a universal rank or a percentage of the market.
Measure the inquiries that matter
A free 30-minute consultation with the owner can review attribution, contact delivery, and which existing pages deserve attention.
Or skip the reading: we will rebuild your homepage free, so you can see the work before you spend anything. You keep the concept either way, wherever you are in the US.
Free and no obligation. Prefer to talk first? Call or text (407) 694-2055. Orlando-based, working with local service businesses nationwide since 2008.