Measuring GEO Results: Separate AI Visibility, Website Visits, and Leads
Three questions, three measures
An AI answer can mention your business without sending a visitor. A visitor can arrive without becoming a lead. A lead can become a customer after several different interactions. Measure each stage separately so your report answers three questions: where is the business visible, which visits can be attributed, and which leads become customers?
No single visibility score answers all three. Start with a recorded baseline and use the same definitions in later reports. Do not assign a dollar value to a citation or form submission without evidence from the business.
Measure identifiable visits in GA4
Google's documented default channels include AI Assistant for visits from recognized AI sources. It excludes Google's AI Overviews and AI Mode, which remain part of Organic Search. Review Session default channel group alongside Session source / medium and landing pages. Check which sources are actually present before introducing custom reporting rules. Google Analytics channel definitions
OpenAI says ChatGPT search referral URLs include utm_source=chatgpt.com. Preserve that parameter through redirects. A citation without a click creates no website visit to measure, and consent choices or missing attribution can limit recorded traffic. An unattributed visit should remain unattributed; do not relabel all direct traffic as AI. OpenAI publisher FAQ
A custom channel group can help inspect identified sources that need a separate reporting view. Document its conditions and order: GA4 assigns traffic to the first matching channel in that group. Do not classify all bing.com or google.com traffic as AI, and do not present a custom group as a complete count of AI influence. Custom channel groups
Track leads and outcomes
Count a form conversion only after the submission succeeds. Test that one successful submission produces one event, then compare analytics totals with saved lead records. Keep names, email addresses, phone numbers, and message text out of analytics event parameters.
For calls, a recorded intake answer can add useful context. Ask how the caller heard about the business and store the answer in the lead system. Treat it as self-reported attribution. Keep qualified leads, booked jobs, and realized revenue distinct so a form submission is never presented as a completed sale.
Read visibility reports separately
Use verified Google Search Console and Bing Webmaster Tools properties to inspect crawl access, indexing, and available search or AI reports. Record the date range and the definition of each metric. A processed sitemap shows discovery; it does not establish that every page is indexed or cited.
Google describes its generative-search optimization as part of SEO. Bing's AI Performance report provides another view of citation activity. Neither is a complete cross-platform census of every AI answer. Consult each platform's current reporting guidance and state its coverage alongside results. Google guidance, Bing AI Performance
Use manual prompts as a diagnostic sample
A repeatable prompt log can reveal examples worth investigating. Record the exact prompt, date, platform, relevant location or account context, answer, cited URL, and whether the business was merely mentioned or linked. Avoid entering private customer information.
If the business appears in eight of twenty recorded answers, report that observed sample result. It is not a market-wide visibility share. A change in a small sample does not establish that one website edit caused it; answers, platform behavior, and competing sources can also change. Do not diagnose a missing recommendation as a schema error without evidence.
Build a report that can guide action
Use a consistent date range and show citations or mentions, attributable sessions, successful lead events, qualified leads, and booked jobs as separate measures. Include the source, definition, and known gaps for each. Compare periods only when collection and definitions are sufficiently consistent, and annotate changes to tracking or consent.
Investigate the stage where evidence shows a problem. Crawl failures require access checks. Incorrect business information requires factual correction. Visits without qualified leads call for a review of audience fit and the customer journey. Choose the next action from observed evidence rather than a promised citation score or an unsupported revenue benchmark.