How to measure AI referrals, enquiries and booked calls

Measure AI visibility and commercial results in separate records, then connect them where the evidence allows. A citation does not prove a visit. A visit does not prove an enquiry. A click on “book a call” does not prove that a booking was completed.

For a small business, a simple analytics report and a maintained enquiry log can be more useful than a complicated visibility score. Agree the definitions before making changes so the eventual report answers whether the work helped the business.

Use six measures with different meanings

Set up the visit evidence

In your analytics tool, inspect the landing page, source, medium and available campaign information. Google’s traffic-source documentation explains how campaign values and referrers are collected. Keep the reporting scope consistent: a session source and a person’s first recorded acquisition source answer different questions.

Maintain a small, reviewed list of AI referral sources actually observed in your data. Keep the original source value alongside any grouped label. Do not classify every Google or Bing visit as AI traffic, or assume that an unfamiliar referrer belongs to an AI product.

Test a real journey through the website using an agreed test marker: landing page, enquiry form, confirmation and booking. Check redirects preserve relevant campaign parameters. Keep internal navigation links free of acquisition tags that would muddy the journey. Apply the site’s consent choices throughout, and keep personal form contents out of analytics events.

Unknown attribution must remain unknown. Google describes direct traffic as traffic without a clear referral source; several different causes can produce it. An increase in direct visits is not evidence that AI caused the increase.

Connect analytics to enquiries

Give each legitimate enquiry a reference in the customer system. Record received date, landing page where known, observed acquisition evidence, service requested, qualification outcome and booking status. Avoid counting repeat submissions as separate prospects.

Add an optional “How did you hear about us?” question if it fits the form. Preserve the answer as self-reported evidence. Someone saying “ChatGPT” is useful information, but it is a different measurement from a captured referral and should not overwrite it.

Agree what qualified means. For example, a service business might require a relevant need, a territory it serves and willingness to discuss a realistic project. Keep those criteria stable during the reporting period; relaxing them would make growth look better without improving demand.

Keep platform reports alongside the funnel

Google’s Generative AI performance report provides AI Overviews and AI Mode impressions. It does not by itself establish which enquiry came from an AI answer. Its activity also overlaps overall Web Search reporting, so avoid adding overlapping totals.

Bing AI Performance reports citations across supported surfaces using sampled and aggregated data. Grounding queries are retrieval phrases, not a transcript of individual prompts. Citation share is not traffic share, and missing low-frequency data should not be reported as proof of absence.

Save manual observations separately using the AI visibility tracking template. Keep the question set, platform and location context attached to any percentage you report.

A worked example, not a forecast

Suppose a fictional monthly report records 20 attributable AI referral sessions, two distinct received enquiries associated with those sessions and one completed booking. The session-to-enquiry ratio is 2 ÷ 20, or 10%. The enquiry-to-booking ratio is 1 ÷ 2, or 50%.

Those small denominators matter. One extra booking changes the second ratio dramatically. Neither figure is a forecast, and neither includes people whose source could not be identified. If one of the two enquiries came only from self-report, keep it in that separate category rather than silently changing the denominator.

Make the monthly decision

Review comparable periods, note tracking failures and list material website or campaign changes. Look for questions generating useful conversations, pages producing relevant enquiries and gaps that block booking. Investigate a pattern before attributing it to one edit.

Start with the AI visibility audit guide if the baseline is missing. For Google-specific eligibility, use the AI Overviews and AI Mode checklist. Our AEO service can agree the measurement definitions and priority fixes before any ongoing work begins.

Official guidance checked on 27 October 2026. The numerical example is illustrative.