How do I measure customers from ChatGPT?
Count a ChatGPT-referred customer only when you have both source evidence and a defined customer outcome. An answer mentioning you, a citation, a visit, a signup, and a payment are different events. Retain the available referrer or campaign tags, connect them to the inquiry or account through your approved analytics system, and count paid customers from confirmed payment records. Keep survey answers separate and label missing attribution unknown.
Keep a funnel with explicit definitions
Define what each step means before launching content. For a service business, first value might be a completed scan or a useful diagnostic request; for a product, it might be completing the job the tool exists to do. Neither is automatically a paying customer.
- Visibility: a dated assistant answer mentions or cites a page.
- Referral visit: an identifiable visit arrives from an assistant or carries a recognized source tag.
- Qualified inquiry: the request fits your stated service and scope.
- Activation: the visitor completes the useful action you defined.
- Paid customer: an authoritative payment record confirms the agreed purchase.
Retain source evidence without collecting the conversation
Use the analytics and consent controls already approved for your site. Record the landing page, available referral host, recognized campaign values, and conversion timestamp. Associate account or request records only where your privacy policy and access controls permit it. You do not need to collect the buyer's private chat transcript.
- Preserve first-touch and conversion-touch sources separately when supported.
- Keep “How did you find us?” as a separate self-reported field. Reconcile overlaps; do not add it to tracked referrals as if the same person were two customers.
- Missing referrers remain unknown. Campaign tags, when present, indicate attribution, not a platform endorsement or proof of incremental growth.
- Exclude controlled tests and internal agent visits from business results.
Run a small comparison before scaling
Record a baseline, publication dates, and other promotions or product changes. Improve a small set of pages and compare them with similar unchanged pages where possible. Check a fixed set of buyer questions weekly, using repeated fresh conversations. Report actual visits, qualified actions, and payments alongside channel shares. If indexing is slow or counts are small, extend the observation window rather than announcing a winner. A before-and-after increase alone does not isolate the effect of the content.
Sources and further reading
- OpenAI: ChatGPT Search — How web search and citations work; placement is not guaranteed.
- ChatGPT recommendation readiness — What you can control before measuring referrals.
Where to go from here
The free checks come first — run them before paying anyone, including us.
Nymrel AI-visibility scan
Free, no email or signup — eight technical checks on any URL, scored in about a minute.
Download the referral measurement worksheet
Free CSV headers for your own baseline and weekly counts. It contains no customer data or example results.
Review the website diagnostic
Read the current scope and pricing before requesting help. Implementation and optional monitoring are separate.
Related questions
Does half of signups mean half of paying customers?
No. The denominator matters. Report signups, activated users, and paid customers separately, with the exact period and attribution method for each.
Does an assistant citation mean someone visited my site?
No. A citation is a visibility observation. A visit requires traffic evidence, and a customer requires the business outcome you defined.
Can a 30-day pilot prove the strategy caused growth?
Not by itself. Similar unchanged pages and a stable measurement method improve the comparison, but small samples, indexing delays, shared-domain effects, and other changes can limit causal conclusions.