Measurement
Measure AI Search Leads from Visit to Sales Review
Track identifiable AI referrals, remove duplicate inquiries and define qualified leads with sales. Includes a worked attribution example.
TL;DR
- Define a qualified inquiry with sales before reporting AI-sourced leads.
- Join visits and inquiries only where the evidence supports it; keep self-reported discovery distinct.
- Test the handoff, deduplicate records and preserve pending or unknown outcomes.
A screenshot of your business in an AI answer is an observation, not a sales result. It can help you understand how a platform presents your company, but it does not tell you how many relevant buyers saw it, visited your website, or requested a proposal.
The plan should link what AI answers show, what happens on the site and which sales requests are a fit. Keep each stage clear in the report. This guide shows a method a B2B team can use; it does not claim that all visits can be traced.
Use an inquiry ledger to resolve attribution gaps
Adobe’s March 2026 US retail analysis reported AI-referral conversion 42% higher than non-AI traffic, time on site 48% longer and pages per visit 13% higher. Those relative figures come from retail data, not a B2B sales funnel. They cannot supply your lead target.
Worked lead review: four records, three people
This is a made-up example. The review uses a consent-appropriate analytics export, form records and sales status notes. The issue is duplicate submissions and mixed source evidence.
| Record | Evidence | Treatment |
|---|---|---|
| A: buyer requests a relevant service | Identified AI-referral session and matching inquiry; sales confirms fit. | One qualified AI-referral inquiry. |
| A again: same buyer retries the form | Same person and request; duplicate delivery. | Merge into A. Do not count a second lead. |
| B: buyer says an AI tool helped them | Self-report, but no identifiable AI referral for the visit. | One self-reported AI-assisted inquiry; keep separate from observed referrals. |
| C: identifiable AI referral, unclear need | Valid inquiry; sales has not reviewed fit. | One AI-referral inquiry, qualification pending. |
The output is three unique inquiries, of which two have AI referrals with a known source. Only one is confirmed qualified; one is pending and B uses a different evidence label. Do not divide the three inquiries by only the AI-referral sessions. That would mix sources in the numerator and denominator.
The review is done when sales can trace each status to one request, the analyst can explain each source label, and a test reaches the right owner. Use an authorized test setup; do not submit fabricated buyer details to a live sales team. The reporting template shows how to carry these rules into a monthly report.
Define a qualified inquiry first
Agree with sales what makes a request a fit. Use clear rules for service need, market, type of business and the problem you can solve. Two people should reach the same verdict on the same message.
Track the stages your team uses: received, qualified, meeting held, proposal requested and outcome. A form submission is not a qualified deal just because it came from search.
Establish a baseline you can explain
Save the search and inquiry data before changing key pages. Use a period that fits your sales cycle and traffic. A site with few leads may need a longer window to show a useful pattern.
Keep a dated change log beside the report. Note launches, page edits, campaigns, tracking changes and events in the sales process. If a new page and an ad campaign launch together, do not credit all new leads to the page.
Show gaps too. “Tracking was unavailable” is more useful than a zero that looks measured.
Understand the limits of search reporting
Google’s Generative AI performance report shows impressions for AI Overviews and AI Mode. These are not site visits or sales requests. A report with too little data may be absent. Keep an unavailable report apart from a measured zero.
Do not call all Web clicks AI-driven visits. Report what you can trace and label any inference. Place platform data, site sessions and inquiries beside one another, but join them only where the records support it.
Use a repeatable observation sample
Choose a fixed set of real buyer questions. Record the platform, exact wording, date, location or account state where relevant, answer and sources. A brand mention differs from a link to your site.
Repeat under the same stated conditions. Keep missing mentions and wrong claims as well as good answers. One sample does not measure the whole market. Ask a provider how its score handles failed or changing answers.
Use the review to choose a task: clarify scope, gather missing proof or check a code issue.
Check the handoff into sales
In an authorized test, check that a valid request is saved, reaches its owner and gives the visitor an accurate success message. Test a failed send and a duplicate too.
Where tools and privacy choices allow, retain the landing page and source context. A voluntary “How did you hear about us?” answer can help, but it is not a full record of the buyer’s research.
Review a small set of real inquiries with sales each month. Ask which pages helped and which questions remained. That can guide the next edit even when the source is unknown.
Report a decision, not just a dashboard
End with what changed, what is unclear and who will do what next. Check weak points before adding more pages. The GEO and SEO planning guide can help choose a useful first phase.
When you request a proposal, include the site and how your team receives and checks sales requests.
Frequently asked questions
Can a buyer mention AI after arriving through another channel?
Yes. Discovery can span several visits and channels. Keep the self-report without rewriting the observed session source.
Does an unknown source mean the inquiry has no value?
No. Source evidence and sales value are separate fields. Sales can qualify an inquiry even when attribution is unknown.
Should pending inquiries count as qualified?
No. Keep a pending status and review it later. Otherwise changes in review speed can look like changes in lead quality.
Can this prove incremental revenue from GEO?
Not on its own. It links evidence you have to inquiries. Incremental revenue needs a stronger design, a cost basis and a suitable sales window.
Share your inquiry fields and qualification rules to scope the measurement and handoff work.
